- Email: [email protected]
Achieving quality assurance functionality in the food industry using a hybrid case-based reasoning and fuzzy logic approach
Expert Systems with Applications 39 (2012) 5251–5261
Contents lists available at SciVerse ScienceDirect
Expert Systems with Applications journal homepage: www.elsevier.com/locate/eswa
Achieving quality assurance functionality in the food industry using a hybrid case-based reasoning and fuzzy logic approach S.I. Lao a, K.L. Choy a,⇑, G.T.S. Ho a, Richard C.M. Yam b,1, Y.C. Tsim a, T.C. Poon a a b
Department of Industrial and System Engineering, The Hong Kong Polytechnic University, Hunghom, Hong Kong Department of System Engineering & Engineering Management, The City University of Hong Kong, Hong Kong
a r t i c l e
i n f o
Keywords: Food quality Case-based Reasoning Fuzzy logic Decision support system Operation guidelines Storage conditions
a b s t r a c t Quality control of food inventories in the warehouse is complex as well as challenging due to the fact that food can easily deteriorate. Currently, this difficult storage problem is managed mostly by using a human dependent quality assurance and decision making process. This has however, occasionally led to unimaginative, arduous and inconsistent decisions due to the injection of subjective human intervention into the process. Therefore, it could be said that current practice is not powerful enough to support high-quality inventory management. In this paper, the development of an integrative prototype decision support system, namely, Intelligent Food Quality Assurance System (IFQAS) is described which will assist the process by automating the human based decision making process in the quality control of food storage. The system, which is composed of a Case-based Reasoning (CBR) engine and a Fuzzy rule-based Reasoning (FBR) engine, starts with the receipt of incoming food inventory. With the CBR engine, certain quality assurance operations can be suggested based on the attributes of the food received. Further of this, the FBR engine can make suggestions on the optimal storage conditions of inventory by systematically evaluating the food conditions when the food is receiving. With the assistance of the system, a holistic monitoring in quality control of the receiving operations and the storage conditions of the food in the warehouse can be performed. It provides consistent and systematic Quality Assurance Guidelines for quality control which leads to improvement in the level of customer satisfaction and minimization of the defective rate. Ó 2011 Elsevier Ltd. All rights reserved.
1. Introduction Food quality is an important issue in the food industry. Lots of quality checking and assurance duties are required throughout the whole food chain. Poor quality control decisions may lead to a high level of defective goods and a poor level of customer satisfaction (Nilsson, Johnson, & Gustafsson, 2001; Youngdahl & Kellogg, 1997). Currently, food safety assuring systems like Hazard Analysis and the Critical Control Point (HACCP) system are widely promoted for demonstrating the commitment to food safety in the food industry (Orriss & Whitehead, 2000). The warehouse, where food is stored and where value adding activities are performed, is no exception to this commitment. In order to ensure the food is of acceptable quality, appropriate quality control actions need to be performed (Getinet, Seyoum, & Woldetsadik, 2008; Yan, Sousa-Gallagher, & Oliveira, 2008). However, the operations and duties that need to be undertaken for the sake of safety assurance are complex and difficult to apply.
⇑ Corresponding author. E-mail addresses: [email protected] (K.L. Choy), [email protected] (R.C.M. Yam). 1 Tel.: +852 34428417; fax: +852 34420172. 0957-4174/$ - see front matter Ó 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.eswa.2011.11.014
In fact, the current quality assurance process adopted in warehouses has several serious problems. Traditionally, the decision making process for selecting the necessitate quality control operation relies mainly on the skills and experience of operators. This means that errors can easily occur. Nevertheless, the increasing niche requirements of diversified value added activities and the great variance in Stock Keeping Unit (SKU) which require entirely different handling operations in the warehouse, further increase the complexity of the quality control. On top of this, numerous researchers have undertaken different studies in promoting and evaluating the importance of adopting various quality assurance systems concerned with food handling. However, the actual practical automation of a quality control assistance system in warehouses, as a research field, has not yet been much explored. An investigation into the adoption of a decision support system (DSS) for automating the process may help to improve the situation. The purpose of this paper is to outline and illustrate a decision support approach to automate the existing human based decision making practice for determining the appropriate quality assurance operations for food inventory management. An integrative prototype system, namely, the Intelligent Food Quality Assurance System (IFQAS), has been developed, not only for facilitating the selection of the most appropriate quality control operations, but also for suggesting the best storage environment for the goods after the quality
5252
S.I. Lao et al. / Expert Systems with Applications 39 (2012) 5251–5261
has been checked. In the system, a Case-based Reasoning (CBR) engine which solves new cases by reusing the previous handling experience in decision making is proposed to replace the manual decision making patterns in deciding the quality assurance operations. Because of the special nature of the food items it is not possible for humans to judge accurately the condition of the food inventory or express the conditions in precise numerical values, so often vague, fuzzy logic techniques are applied in order to extract the critical quality assurance information in terms of fuzzy rules. Therefore, a Fuzzy rule-based Reasoning (FBR) engine is constructed for suggesting the appropriate storage environment. The rest of this paper is organized as follows. In Section 2, the current quality assurance method used in the food industry with the application of various kinds of technology is reviewed. In Section 3, the system architecture of an Intelligent Food Quality Assurance System (IFQAS) and the mechanism of its two modules are described and explained in detail. In Section 4, a case study for validating the feasibility of the adoption of IFQAS is presented. Section 5 contains a detailed discussion of the system’s performance. The final section, Section 6, concludes the paper. 2. Related works Deterioration in food quality may lead to a huge impact on the credibility of a food company (Shapton & Shapton, 1991) and products that have been subjected to a compromise of different conditions and sensory properties would affect customer’s choice and preference during purchasing (Perrot et al., 2004). Now, there is a movement afoot to change the current human-judgment based food quality control approach to a more scientific and systematic one. When the quality and conditions of food are solely interpreted by the human operators, errors occur easily (Perrot et al., 2006). Human based quality assurance or decision making can manage the operations, but are believed to be unimaginative and arduous and lead to inconsistent actions (Davidson, Ryks, & Chu, 2001). Therefore, solely relying on human observation and evaluation is not systematic or reliable enough. This indicates that implementing quality assurance or control measurement is a critical issue. Researchers have explored this issue, however, it was found that their emphasis was on the importance of implementing such assurance system (Blaha, 1999), investigating the relationship between factors, such as the consequence of poor quality, the quality culture and the effectiveness of the goals achieved and the activities which lead to its adoption, (Karipidis, Athanassiadis, Aggelopoulos, & Giompliakis, 2009), and the provision of an implementation plan (Asefa et al., 2011). Researchers rarely evaluated the decision making process used for selecting the appropriate quality control operation procedures for particular types of food. In addition to this, researchers have adopted different DSS when undertaking quality control operations. Deslandres and Pierreval (1997) have designed a knowledge advisory system for quality applications; a quality advisory model is built for structuring the knowledge required for quality problem solving. Carpenter and Maropoulos (2000) have developed a system called OPTIMUM for controlling the tool machining process. It makes use of a combination of mathematical modeling and rule-based statistical methods for decision making. Shaffer and Brodahl (1998) have proposed a rule-based management system for offering management solutions for maintaining better farming conditions. Their predefined rules are stored in a database for decision making. However, those systems lack the self-learning capability of being able to procure human being’s knowledge. In handling food related operations, the decision making process involves numerous adjustments, therefore, self-learning and improving are important. As it has this special skill it is believed that Case-based Reasoning (CBR) is probably the right choice. CBR is one of the most popular
techniques for solving problems in the selection of operations. Peng, Chen, Wu, Xin, and Jing (2011) have developed a virtual reality based integrated system, applying CBR for matching the most similar machining fixture design case from a case-based database. It helps assign the production operations in the designing of parts. Kalapanidas and Avouris (2001) have proposed a NEMO prototype for improving the quality of air in operating conditions. It combines heuristic and statistical techniques based on the CBR approach for suggesting solutions. Chow, Choy, Lee, and Lau (2006) have applies CBR for selecting the most suitable resource usage package for handling warehouse operations. It found that using CBR engine for suggesting the operations is time saving and cost effective. Xia and Rao (1999) adopted CBR technology in an operation support system. The implementation result pointed out that CBR helps achieve more consistent and accurate operations. It found that there has been a trend to adopt learning techniques which use cases or instances directly, such as in Case-based Reasoning (CBR) (Kim, Im, & Park, 2010). In general, it is believed that CBR is particularly suitable for solving problems of domains that are experience-rich but knowledge poor (Chi, Chen, & Kiang, 1993). Nevertheless, the CBR techniques may only be capable of handling the decision making process at the operations selection level, but it may not be applicable for solving problems in food handling such as monitoring storage conditions. To reach a highquality level and identify errors in operations, attention should be paid to all of the characteristics of the sub-processes (Guillaume & Charnomordic, 2001). Probably, a system which can provide an all-round quality assurance support down to the parameter level is required. With reference to the published studies, Pacella and Semeraro (2005) applied Artificial Neural Networks (ANN) to the management of quality control in manufacturing processes. It recalled the learning patterns from the incomplete representations in order to monitor the quality characteristics. The system helps detect, classify and predict any unnatural changes that occur in the manufacturing process. Bezerra et al. (2007) have used ANN for data classification and pattern recognition for predicting material behavior for quality prediction. Jung and Yum (2011) have proposed an ANN-based approach for mapping the relationship between the characteristics, design and signals for taguchi parameter design. However, ANN need their inputs to be expressed in numeric terms for the decision making to process (Metaxiotis, Ergazakis, Samouilidis, & Psarras, 2003), the relationship between the status and the condition of the food items as the storage environment evaluation is complex and difficult to describe through modeling. Instead of using modeling, tackling this relationship using fuzzy logic approach is a probable solution. Fuzzy logic provides a methodology that managing blurry attributes and allow the use of data and information from those who possess expert knowledge. It has become an increasing important approach for tackling food problems and handles human reasoning in linguistic terms (Guillaume & Charnomordic, 2001). It is found that fuzzy logic techniques have been widely used in managing food problems. Researchers (Ioannou, Perrot, Curt, Mauris, & Trystram, 2004; Lababidi & Baker, 2003; Perrot et al., 2004) have adopted the fuzzy logic in assisting the quality control of food items during the production process. It is found that the concept and terms describing the quality of food items, and the justifications in the human mind are in an area of uncertainty and vagueness. Therefore, it is difficult to adjust and make decisions based on the measurement results in terms of numbering and crisp values. Fuzzy logic is considered as a tool that is suitable for dealing with fuzzy relationships, criteria and phenomena (Amelia, Wahab, & Hassan, 2009; Lin & Hsieh, 2004). Applying fuzzy terms in decision making, helps capture the reasoning process or the hidden uncertainty of operations (Jiang & Chen, 2005). Hence, the development of fuzzy sets helps build the linkage between words and numbers.
S.I. Lao et al. / Expert Systems with Applications 39 (2012) 5251–5261
5253
Fig. 1. Operation flow of the Intelligent Food Quality Assurance System.
Despite the wide application of fuzzy logic in food production monitoring, the technique has yet to be applied in the area of food safety assurance. Therefore, fuzzy logic is proposed in this paper for suggesting the storage conditions in warehouses. After examining the special needs and characteristics of food, this study attempts to propose a DSS using CBR and fuzzy logic to determine the appropriate quality assurance operations for food inventory management.
Hence, other than operation guidelines, a set of suggested storage environmental conditions is generated with the help of the fuzzy inference module by evaluating the food conditions. These suggestions further provide monitoring of the conditions down to the parameter level. In Fig. 2, the system architecture is shown. The Intelligent Food Quality Assurance System (IFQAS) consists of three tiers: the system database module, CBR module, and FBR module. The detailed operation mechanism of the system is described as follows:
3. The Intelligent Food Quality Assurance System (IFQAS) In this section, the Intelligent Food Quality Assurance System (IFQAS) is presented that provides knowledge based decision support for food management. The proposed IFQAS controls the food quality assurance operations in a warehouse by gathering different food attributes within the receiving and storage process, and converting the data into knowledge in terms of case and fuzzy rules. Quality assurance operations and the optimal storage environment are suggested based on the knowledge stored in the knowledge domain. As shown in Fig. 1, order specifications of the inbound inventory which are gathered from the workplace and the company database are collected. With the assistance of the CBR engine, suggested operation guidelines for quality assurance are generated.
3.1. Database module In the database module, different inventory parameters are stored. These data are grouped into two types: Order Specifications and Food Conditions. Order Specifications consist of physical information that is specified in the order invoices and on packaging, such as location of origin, quantity, storage type, and commodity name. This information is required for visualizing the categories and requirements of the foods. Information regarding Food Conditions is related to the environmental conditions and quality of the various foods. These data include temperature, color, elasticity and product size.
Fig. 2. System architecture of Intelligent Food Quality Assurance System.
5254
S.I. Lao et al. / Expert Systems with Applications 39 (2012) 5251–5261
3.2. CBR module In this tier, a CBR module is constructed to provide suggestions for the quality assurance operations. The CBR engine retrieves information about the inventories and about the existing quality standards of the company. The data act as the attributes for the retrieval of historical cases. Cases with the highest similarity value to the new case would be adopted. Modification may take place to enhance the case suitability. There are four phases in the CBR module: (i) Retrieve, (ii) Reuse, (iii) Revise, and (iv) Retain.
Fig. 3. IFQAS implementation steps.
(i) Retrieve Previous records are treated as cases and are divided into problem and solution parts. The problem part includes relevant case features, and the solution part mentions handling conclusion and comments. When food arrives, the CBR engine would activates the system to browse for the required information for the generation of key attributes. Using the list of generated key attributes, the CBR engine searches for potentially useful cases in the case library. Inductive indexing is applied for clustering cases with various similarities to the case being handled. Hence, a group of potentially useful cases in the resulting cluster with the tree structured searching, are generated and presented for case reuse. (ii) Reuse With the potential cases list generated, similarity values of the potential cases to the newly cases are calculated based on Eq. (1). The total similarity value is calculated base on the summing up of the similarity value of all attributes with weightings. With reference to all the total similarity values, cases are ranked in descending order:
Pn
I R i¼1 wi simðfi ; fi Þ Pn i¼1 wi
on the food safety during storage. Based on the food conditions, an appropriate storage environment is suggested, so as to help maintain the food quality. The controlling parameter is storage temperature. The fuzzy inference engine is composed of a rule-based reasoning engine that using fuzzy logic for solution suggestions. Generally, there are three components in the fuzzy inference: a fuzzifier for translating the crisp inputs to fuzzy values; an inference engine for applying a fuzzy reasoning mechanism to obtain a fuzzy output and a defuzzifier for translating a fuzzy output into a crisp value. Besides these, there is a knowledge base which contains fuzzy rules in terms of a rule-base and membership functions as a database. The characteristics of the fuzzy set are determined by membership functions. The present conditions of the food inventories need to be gathered so that they can act as the measurement data of the engine. Those data are then transformed into the right format, and act as the input data set of the fuzzy system. Knowledge from the workers and the implicit recording of human decisions and process are extracted and then analyzed. This knowledge is then turned into IF-THEN rules and stored into the rule set database for decision making. Those rules are part of the fuzzy system.
ð1Þ
where wi is the weight of attribute I and fiI ; fiR are the value of fi in the input. The case ranked with the highest similarity value is treated as the most significant case for action suggestion. The final operations suggestion would mainly consider the most highly ranked case. Cases with comparatively lower total similarity value would only serve as a reference in the revision stage.
3.3.1. Fuzzification Generally, there are four membership functions used in the engine. Fuzzy sets are defined by membership functions and each of the elements is designated as belonging to a degree of membership (between 0 and 1). A control framework is built which makes decisions based on different dimensions of the product quality, such as temperature, color, elasticity and product size. Fuzzy set temperature (T):
(iii) Revise When a historical case is suggested by the CBR engine, modifications, like editing and combining, are carried out if necessary. This step ensures the final case suggestion fits the actual situation and increases the adoptability of the solution. (iv) Retain The generated revised report and the new case content are then sent to the case library for storage. It acts as a reference and case asset for future use. After the CBR engine has been used for a period of time, the case database maturity increases. 3.3. Fuzzy rule-based module In this tier, a fuzzy inference module namely, Fuzzy rule-based module (FRB), is constructed for further processing of the CBR case results. With the generated CBR’s case suggestion, those foods that require temperature control monitoring are identified. In order to further ensure a holistic quality control to be performed, the fuzzy inference module provides expert advice and qualitative prediction
Fig. 4. Operation platform of IFQAS.
S.I. Lao et al. / Expert Systems with Applications 39 (2012) 5251–5261
5255
Fig. 5. System input of the relationship and importance among attributes.
Fig. 7. Quality Assurance Guidelines suggested from the CBR module. Fig. 6. Quality Assurance Guidelines cases suggestion.
T¼
n X
lT ðti Þ
ij1
ti
T, the whole data set; t, element of subset T. Fuzzy set Color (C):
C¼
n X
lC ðci Þ
ij1
ci
C, the whole data set; c, element of subset C. Fuzzy set Elasticity (E):
E¼
n X
lE ðei Þ
ij1
ei
E, the whole data set; e, element of subset E. Fuzzy set Size (S):
S¼
n X
lS ðsi Þ
ij1
si
S, the whole data set; s, element of subset S. In each of the membership functions, different membership values are defined. Those values are named the universe of discourse. It is divided into several regions for prediction of the food conditions. This can be exemplified by using temperature:
Temperature ðTÞ : fH; RH; M; RL; Lg where are high (H), relatively high (RH), medium (M), relatively low (RL), low (L). With the input data, the corresponding membership values of different attributes are calculated. 3.3.2. Inference engine The quality perception mechanism and the relationship between fuzzy input and fuzzy output variables are presented in
5256
S.I. Lao et al. / Expert Systems with Applications 39 (2012) 5251–5261
Fig. 8. The architecture of FBR.
the form of simple and intuitive ‘‘IF-THEN’’ fuzzy rules. The fuzzy rules were developed based on the results of interviewing quality experts, evaluating the industrial data and referring to the literature. The set of rules of the fuzzy relation are formulated, for example: Rule 1: IF A is high AND B is low THEN C is good. Rule 2: IF A is very low AND B is medium THEN C is bad. 3.3.3. Defuzzification To defuzzify the output fuzzy terms, the center of area (COA) is adopted to convert the fuzzy output back to a crisp value. The generating equation is
Pn xi Ai X ¼ Pi¼1 i¼1 Ai
where x is the center of gravity representing the probability of the case occurring; Ai denotes the area for each individual result. By calculating the COA, the fuzzy output is converted into crisp values with reference to the output membership function.
4. Case study In order to validate the effectiveness of the system, a pilot test was undertaken in one of the Hong Kong based Logistics companies. This company mainly handles food related inventories. Salmon fish is one of the core foods that this company handles. When salmon is received, it undergoes quality checking, customerized value added activities, and storing and then the salmon is delivered to the retail shops. Different quality checking and storage procedures are required for salmon which originates from different
Fig. 9. Input membership function (Color Score).
S.I. Lao et al. / Expert Systems with Applications 39 (2012) 5251–5261
5257
Fig. 10. Output membership function (Storage Temperature).
places, and which has been kept under different conditions and which is of different categories. Although quality assurance is required as part of the receiving process, the increase in value added activities increases the variety of the control points required and the difficulties in managing the operations. All of the decision making processes rely on the experience and knowledge of workers. There are no standard rules or regulations. Because of the numerous problems entailed in managing all the receiving operations, a trial run on IFQAS was undertaken in the company in August 2010 for a month. There are four implementation steps in total as illustrated in Fig. 3. 4.1. Data retrieval in the Warehouse Management System (WMS) Initially, the company adopted WMS for managing the warehouse inventory assignment functions. When goods are inbound
into the warehouse, data and information such as the quantity, SKU, the SKU dimensions, commodity name and the storage type are input into the WMS. In order to extract information for the CBR engine, an operation platform is built for data extraction. The operators key in the item number and order number to the system, click the ‘‘NEXT’’ button, and the screen will show the details of the order. This information is important for the evaluation of the inventory quality assurance requirements of the CBR module (Fig. 4). 4.2. Case construction of the CBR engine in IFQAS While constructing the CBR engine, two types of information are required: (i) The specific attributes for decision making. The usual practise of the company when managing the quality assurance operations is to rely on human justification based on the knowledge
Fig. 11. Fuzzy rule set.
5258
S.I. Lao et al. / Expert Systems with Applications 39 (2012) 5251–5261
Table 1 Matlab data code of the input variable ‘‘Color Score’’. [Input1] Name = ‘Color_Score’ Range = [20 34] NumMFs = 5 MF1 = ‘Poor’: ‘trapmf’, [19 20 21 24] MF2 = ‘Slightly_Poor’: ‘trimf’, [21 24 27] MF3 = ‘Medium’: ‘trimf’, [24 27 30] MF4 = ‘Good’: ‘trimf’, [27 30 33] MF5 = ‘Very_Good’: ‘trapmf’, [30 33 34 35]
of the staff about food management and the requirements of different food regulations such as Hazard Analysis and Critical Control Points (HACCP). Therefore, before turning the humanbased decision making process into a case-based decision DSS, the knowledge and the experience of the workers in the quality assurance department of the company is extracted and converted into cases. Several interviews are conducted for converting the human tacit knowledge into cases. Besides, all the previous salmon handling records of the company are studied, restructured and presented as historical cases. Those cases are then stored in the case library with attributes categorized into four levels: Level (1) Storage Types; Level (2) Customer Requirements; Level (3) Operation Specifications; and Level (4) Food Types. (ii) The relationship and importance of the case attributes. By defining the relationship and the importance of attributes, different weightings (wi) can be assigned. Hence, the similarity value (fi) among attributes can be defined. The entity’s importance for decision making and relationship to various attributes are then input into the CBR engine as shown in Fig. 5. With the case information provided by the WMS received by the CBR engine, a number of potentially useful cases are suggested. By calculating the total similarity value (TSV) through the CBR engine with reference to Eq. (1), which is shown in Section 3.2, cases are ranked. Cases with the highest TSV are selected as the most preferred cases for case adoption. As shown in Fig. 6, ‘‘Salmon 0325’’ with a TSV of 95 is selected.
Table 2 Sample rule sets of the system. Rule: 1 IF
(CS) (E) (T) (S)
Color Score of salmon fillet is very good AND Elasticity of meat is elastic AND Temperature of salmon is very low AND Size of salmon is very small
(ST)
Storage Temperature is no change
(CS) (E) (T) (S)
Color Score of salmon fillet is very good AND Elasticity of meat is elastic AND Temperature of salmon is low AND Size of salmon is small
(ST)
Storage Temperature is no change
(CS) (E) (T) (S)
Color Score of salmon fillet is good AND Elasticity of meat is elastic AND Temperature of salmon is very high AND Size of salmon is small
(ST)
Storage Temperature is slightly decreased
(CS) (E) (T) (S)
Color Score of salmon fillet is good AND Elasticity of meat is medium AND Temperature of salmon is medium AND Size of salmon is very small
(ST)
Storage Temperature is slightly decreased
(CS) (E) (T) (S)
Color Score of salmon fillet is slightly poor AND Elasticity of meat is medium AND Temperature of salmon is medium AND Size of salmon is medium
(ST)
Storage Temperature is decreased
(CS) (E) (T) (S)
Color Score of salmon fillet is good AND Elasticity of meat is medium AND Temperature of salmon is high AND Size of salmon is medium
(ST)
Storage Temperature is decreased
(CS) (E) (T) (S)
Color Score of salmon fillet is poor AND Elasticity of meat is inelastic AND Temperature of salmon is very high AND Size of salmon is very large
(ST)
Storage Temperature is heavily decreased
(CS) (E) (T) (S)
Color Score of salmon fillet is very good AND Elasticity of meat is very inelastic AND Temperature of salmon is medium AND Size of salmon is small
(ST)
Storage Temperature is heavily decreased
THEN Rule: 2 IF
THEN Rule: 3 IF
THEN Rule: 4 IF
THEN Rule: 5 IF
THEN Rule: 6 IF
THEN Rule: 7 IF
4.3. Fuzzy inference engine for food safety forecasting
THEN
With the Quality Assurance Guidelines suggested from the CBR module as shown in Fig. 7, the operators are provided with information and guidelines as to what quality assurance operations should be performed and how to perform them upon receipt of the salmon. As salmon belongs to the category that required temperature controls during storage in the warehouse, further operations in storage temperature suggestion is performed. Sensory evaluation is an important method for assessing the freshness and quality of fish. However, fuzzy terms are involved during the analyzing process. Therefore, a fuzzy inference engine is developed for evaluating the initial conditions of the salmon upon receiving it. Based on the conditions of the salmon, optimal storage conditions are suggested. In the past, the quality of the salmon was evaluated by human experts, manually, based on personal observations. Those experts were trained to analyze the condition of the salmon so as to decide the most appropriate storage environment conditions. Basically, the salmon fishes are stored at 4 °C. If the salmon was found to have deteriorated, it will be rejected. However, if the salmon has only slightly degraded, the storage temperature will be adjusted. The aim of making this adjustment is to help lengthen the product
Rule: 8 IF
THEN
life of the salmon by providing an appropriate storage environment. Interviews and evaluation of the company’s historical data was performed in order to convert this tacit and unstructured knowledge and information into rules and facts. By recording the knowledge of handling those processes, hence deducing the hidden methods of food handling, a number of rule sets were formulated. A structured fuzzy knowledge system was constructed using the Fuzzy Logic Toolbox of Matlab (Version 7.6). The structure of the FBR module for storage condition suggestion is shown in Fig. 8. Color, temperature, elasticity and size of salmon products, found to be important quality parameters (Anderson, 2000; Gallart-Jornet, Rustad, Barat, Fito, & Escriche, 2007), are used for
S.I. Lao et al. / Expert Systems with Applications 39 (2012) 5251–5261
5259
Fig. 12. List of fuzzy rules and the linguistic suggestions for storage temperature.
deciding the storage temperature. Those parameters are input into the system for evaluation. 4.3.1. Input membership functions All the four membership functions are defined and input into the system. Fig. 9 shows the membership functions plotted for one of the input variables, ‘‘Color Score’’, as an example. The input variable ‘‘Color Score’’ is a score for reflecting the freshness of salmon with reference to an international color ruler namely SalmoFan. The color of the salmon is one of the most important parameters for accessing its quality. A five point scale is used: Poor (P), Slightly Poor (SP), Medium (M), Good (G), Very Good (VG).
Color ScoreðCSÞ :fP; SP; M; G; VGg Table 1 shows the data code for indicating the difference of different fuzzy regions. Those coding are input into the system. 4.3.2. Output membership function The system has one output linguistic variable; ‘‘Storage Temperature’’ (ST). It has three terms for instructing different corrective actions as shown in Fig. 10. Based on the expert knowledge, the terms are defined and input into the system for indicating the relationship in the membership function. The output membership function ‘‘Storage Temperature’’ is divided into four regions: No change (N), Slightly Decrease (SD), Decrease (D), Heavily Decrease
Storage TemperatureðSTÞ : fN; SD; D; HDg
4.3.3. Fuzzy rules In order to analyze the conditions and quality of the salmon, different rules are set up based on the knowledge of salmon storage, as shown in Fig. 11. Table 2 displaces the sample rule set of the system. By using the rule viewer to input details of the parameters using the graphical expressing diagram as show in Fig. 12, the system automatically suggests the optimal storage condition for salmon storage in the warehouse. In this case, a salmon with a ‘‘Color Score’’ of 28.4, ‘‘Elasticity’’ of 0.762 s, receiving ‘‘Temperature’’ of 1.2 °C and salmon ‘‘Size’’ of 3.86 kg is input for analysis. The system calculates the compatibility of each rule, and then infers a conclusion for each rule. The inferred conclusions from all rules with the same linguistic value are aggregated and combined into a single fuzzy set. Hence, the systems truncating the fuzzy output membership function into numerical value by defuzzification. Finally, a storage temperature of 1.25 °C is suggested. 5. Results and discussion After the pilot test of IFQAS, the results showing the effectiveness of the system are analyzed. The performance of both the traditional, human based approach and the DSS assistance approach with IFQAS are compared in the following section. Two aspects of performance are examined; they are (1) improvement in quality control and (2) improvement in operation efficiency.
5260
S.I. Lao et al. / Expert Systems with Applications 39 (2012) 5251–5261
5.1. Quality control improvement
6. Conclusions
As shown in the case study section, operation guidelines are suggested for all inbound inventories and the storage conditions are further suggested according to the type of inventory that is under temperature control. The operation guidelines are userfriendly which assists the operators to master the food quality assurance process in accordance with the customer requirements, value-added activities and the inventories specifications. By analyzing the data with the CBR engine, the operations that need to be performed for the purpose of quality assurance are shortlisted. The FBR engine then follows the criteria to further suggest the judgment solution for maintaining the inventory quality. Operators can perform specific actions according to the determined guidelines and adjustments. Before this system was implemented, a chaotic manual decision making approach was adopted, which led to errors and inconsistent decisions. IFQAS help to provide holistic quality control of the receiving operations down to parameter levels which improved the inventory defective rate and the customer satisfaction level, as shown in Table 3.
In a warehouse, quality control of inbound inventory is critical. Receiving operations act as the first gate for quality assurance especially with food inventory. Burdensome procedures and regulations make it difficult for operators to carry out their duties efficiently. Hence, the traditional methods mainly considered the quality control activities at the operations level, they rarely focused on improvement at the parameter level. This paper provides a holistic monitoring approach from the receiving docks to the inventory storage stage. It demonstrated a decision support approach to automate the human based decision making process in the determination of quality assurance operations and storage conditions suggestion of inventory at the parameter level. With the support of the decision support technologies like CBR and FBR, a system, namely IFQAS, is designed to assist and automate the safety control decision making process. The aim of the system is to provide consistent and systematic quality assurance, with improvement in customer satisfaction level and in the defective rate. Despite the contributions of the proposed system in the food quality control perspective, the system is not without constraints. The tacit knowledge and operation practices of human experts are computerized for the preservation of past cases and to help in the construction of rules. In order to collect information for those formations, investigation or analytical tasks need to be executed. Such requirements may not be easy for small or immature companies to fulfill, which increases the difficulty of successful adoption. Thus, the generalization of the storage condition suggestions to other warehouse with different food inventories may be limited, as different parameters may need to be analyzed when some special food items are involved. Further investigation on the appropriateness of integrating the Radio Frequency Identification (RFID) technique for automating the process is suggested. RFID can help improve the visibility of the location and status of the inventory and this can further develop a holistic monitoring of the inventory.
5.2. Operation efficiency improvement The human-based decision making process involved manual analyzing, normally manual review on the inventory specifications and relevant quality guidelines, regulations and documents. The manual evaluation process usually required over 15 min, an even longer time frame was required if a new SKU is received. The IFQAS makes use of the data in the WMS by transferring them to the DSS, which reduces the data gathering time. Hence, the system supported decision making process can generate the operation guidelines and storage conditions automatically in a shorter timeframe, as shown in Table 4. As food inventory needs a short timeframe and fast supply chain for handling, the decrease in operation time helps improve the performance.
Table 3 Improvement in customer satisfaction and quality.
Table 4 Time reduction in operation handling.
S.I. Lao et al. / Expert Systems with Applications 39 (2012) 5251–5261
Acknowledgements The authors wish to thank the Research Office of the Hong Kong Polytechnic University for supporting the project (Project Code: RPBK). References Amelia, L., Wahab, D. A., & Hassan, A. (2009). Modeling of palm oil production using fuzzy expert system. Expert Systems with Applications, 36, 8735–8749. Anderson, S. (2000). Salmon color and the consumer. International Institute of Fisheries Economics and Trade, 1–3. Asefa, D. T., Kure, C. F., Gjerde, R. O., Langsrud, S., Omer, M. K., Nesbajjen, T., et al. (2011). A HACCP plan for mycotoxigenic hazards associated with Dry-cured meat production processes. Food Control, 22, 831–837. Bezerra, E. M., Ancelotti, A. C., Pardini, L. C., Rocco, J. A. F. F., Iha, K., & Ribeiro, CH. C. (2007). Artificial neural networks applied to epoxy composites reinforced with carbon and E-glass fibers: Analysis of the shear mechanical properties. Materials Science and Engineering, 464, 177–185. Blaha, Th. (1999). Epidemiology and quality assurance application to food safety. Preventive Veterinary Medicine, 39, 81–92. Carpenter, I. D., & Maropoulos, P. G. (2000). A flexible tool selection decision support system for milling operations. Journal of Materials Processing Technology, 107, 143–152. Chi, R. T., Chen, M., & Kiang, M. Y. (1993). Generalized case-based reasoning system for portfolio management. Expert Systems with Applications, 6, 67–76. Chow, H. K. H., Choy, K. L., Lee, W. B., & Lau, K. C. (2006). Design of a RFID case-based resource management system for warehouse operations. Expert Systems with Applications, 30, 561–576. Davidson, V. J., Ryks, J., & Chu, T. (2001). Fuzzy models to predict consumer ratings for biscuits based on digital image features. IEEE Transactions on Fuzzy Systems, 9(1), 62–67. Deslandres, V., & Pierreval, H. (1997). Knowledge acquisition issues in the design of decision support systems in quality control. European Journal of Operational Research, 103, 296–311. Gallart-Jornet, L., Rustad, T., Barat, J. M., Fito, P., & Escriche, I. (2007). Effect of superchilled storage on the freshness and salting behavior of Atlantic salmon (Salmo salar) fillets. Food Chemistry, 103, 1268–1281. Getinet, H., Seyoum, T., & Woldetsadik, K. (2008). The effect of cultivar, maturity stage and storage environment on quality of tomatoes. Journal of Food Engineering, 87, 467–478. Guillaume, S., & Charnomordic, B. (2001). Knowledge discovery for control purposes in food industry databases. Fuzzy Sets and Systems, 122, 487–497. Ioannou, I., Perrot, N., Curt, C., Mauris, G., & Trystram, G. (2004). Development of a control system using the fuzzy set theory applied to a browning process—A fuzzy symbolic approach for the measurement of product browning: Development of a diagnosis model—Part I. Journal of Food Engineering, 64(4), 497–506. Jiang, Q., & Chen, C. H. (2005). A multi-dimensional fuzzy decision support strategy. Decision Support Systems, 38, 591–598.
5261
Jung, J. R., & Yum, B. J. (2011). Artificial neural network based approach for dynamic parameter design. Expert Systems with Applications, 38, 504– 510. Kalapanidas, E., & Avouris, N. (2001). Short-term air quality prediction using a casebased classifier. Environmental Modeling & Software, 16, 263–272. Karipidis, P., Athanassiadis, K., Aggelopoulos, S., & Giompliakis, E. (2009). Factors affecting the adoption of quality assurance systems in small food enterprises. Food Control, 20, 93–98. Kim, H. K., Im, K. H., & Park, S. C. (2010). DSS for computer security incident response applying CBR and collaborative response. Expert Systems with Applications, 37, 852–870. Lababidi, H. M. S., & Baker, C. G. J. (2003). Web-based expert system for food dryer selection. Computers and Chemical Engineering, 27, 997–1009. Lin, C., & Hsieh, P. J. (2004). A fuzzy decision support system for strategic portfolio management. Decision Support Systems, 38, 383–398. Metaxiotis, K., Ergazakis, K., Samouilidis, E., & Psarras, J. (2003). Decision support through knowledge management: The role of the artificial intelligence. Information Management & Computer Security, 11(5), 216–221. Nilsson, L., Johnson, M. D., & Gustafsson, A. (2001). The impact of quality practices on customer satisfaction and business results: Product versus service organizations. Journal of Quality Management, 6, 5–27. Orriss, G. D., & Whitehead, A. J. (2000). Hazard analysis and critical control point (HACCP) as a part of an overall quality assurance system in international food trade. Food Control, 11, 345–351. Pacella, M., & Semeraro, Q. (2005). Understanding ART-based neural algorithms as statistical tools for manufacturing process quality control. Engineering Applications of Artificial Intelligence, 18, 645–662. Peng, G., Chen, G., Wu, C., Xin, H., & Jing, Y. (2011). Applying RBR and CBR to develop a VR based integrated system for machining fixture design. Expert System with Applications, 38, 26–38. Perrot, N., Agioux, L., Ioannou, I., Mauris, G., Corrieu, G., & Trystram, G. (2004). Decision support system design using the operator skill to control cheese ripening – Application of the fuzzy symbolic approach. Journal of Food Engineering, 64, 321–333. Perrot, N., Ioannou, I., Allais, I., Curt, C., Hossenlopp, J., & Trystram, G. (2006). Fuzzy concepts applied to food product quality control: A review. Fuzzy Sets and System, 157, 1145–1154. Shaffer, M. J., & Brodahl, M. K. (1998). Rule-based management for simulation in agricultural decision support systems. Computers and Electronics in Agriculture, 21, 135–152. Shapton, D. A., & Shapton, N. F. (1991). Principles and practices for the safe processing of foods. Oxford: Butterworth-Heinemann. Xia, Q., & Rao, M. (1999). Dynamic case-based reasoning for process operation support systems. Engineering Applications of Artificial Intelligence, 12, 343– 361. Yan, Z., Sousa-Gallagher, M. J., & Oliveira, F. A. R. (2008). Identification of critical quality parameters and optimal environment conditions of intermediate moisture content banana during storage. Journal of Food Engineering, 85, 163–172. Youngdahl, W. E., & Kellogg, D. L. (1997). The relationship between service customers’ quality assurance behaviors, satisfaction, and effort: A cost of quality perspective. Journal of Operations Management, 15, 19–32.
Contents lists available at SciVerse ScienceDirect
Expert Systems with Applications journal homepage: www.elsevier.com/locate/eswa
Achieving quality assurance functionality in the food industry using a hybrid case-based reasoning and fuzzy logic approach S.I. Lao a, K.L. Choy a,⇑, G.T.S. Ho a, Richard C.M. Yam b,1, Y.C. Tsim a, T.C. Poon a a b
Department of Industrial and System Engineering, The Hong Kong Polytechnic University, Hunghom, Hong Kong Department of System Engineering & Engineering Management, The City University of Hong Kong, Hong Kong
a r t i c l e
i n f o
Keywords: Food quality Case-based Reasoning Fuzzy logic Decision support system Operation guidelines Storage conditions
a b s t r a c t Quality control of food inventories in the warehouse is complex as well as challenging due to the fact that food can easily deteriorate. Currently, this difficult storage problem is managed mostly by using a human dependent quality assurance and decision making process. This has however, occasionally led to unimaginative, arduous and inconsistent decisions due to the injection of subjective human intervention into the process. Therefore, it could be said that current practice is not powerful enough to support high-quality inventory management. In this paper, the development of an integrative prototype decision support system, namely, Intelligent Food Quality Assurance System (IFQAS) is described which will assist the process by automating the human based decision making process in the quality control of food storage. The system, which is composed of a Case-based Reasoning (CBR) engine and a Fuzzy rule-based Reasoning (FBR) engine, starts with the receipt of incoming food inventory. With the CBR engine, certain quality assurance operations can be suggested based on the attributes of the food received. Further of this, the FBR engine can make suggestions on the optimal storage conditions of inventory by systematically evaluating the food conditions when the food is receiving. With the assistance of the system, a holistic monitoring in quality control of the receiving operations and the storage conditions of the food in the warehouse can be performed. It provides consistent and systematic Quality Assurance Guidelines for quality control which leads to improvement in the level of customer satisfaction and minimization of the defective rate. Ó 2011 Elsevier Ltd. All rights reserved.
1. Introduction Food quality is an important issue in the food industry. Lots of quality checking and assurance duties are required throughout the whole food chain. Poor quality control decisions may lead to a high level of defective goods and a poor level of customer satisfaction (Nilsson, Johnson, & Gustafsson, 2001; Youngdahl & Kellogg, 1997). Currently, food safety assuring systems like Hazard Analysis and the Critical Control Point (HACCP) system are widely promoted for demonstrating the commitment to food safety in the food industry (Orriss & Whitehead, 2000). The warehouse, where food is stored and where value adding activities are performed, is no exception to this commitment. In order to ensure the food is of acceptable quality, appropriate quality control actions need to be performed (Getinet, Seyoum, & Woldetsadik, 2008; Yan, Sousa-Gallagher, & Oliveira, 2008). However, the operations and duties that need to be undertaken for the sake of safety assurance are complex and difficult to apply.
⇑ Corresponding author. E-mail addresses: [email protected] (K.L. Choy), [email protected] (R.C.M. Yam). 1 Tel.: +852 34428417; fax: +852 34420172. 0957-4174/$ - see front matter Ó 2011 Elsevier Ltd. All rights reserved. doi:10.1016/j.eswa.2011.11.014
In fact, the current quality assurance process adopted in warehouses has several serious problems. Traditionally, the decision making process for selecting the necessitate quality control operation relies mainly on the skills and experience of operators. This means that errors can easily occur. Nevertheless, the increasing niche requirements of diversified value added activities and the great variance in Stock Keeping Unit (SKU) which require entirely different handling operations in the warehouse, further increase the complexity of the quality control. On top of this, numerous researchers have undertaken different studies in promoting and evaluating the importance of adopting various quality assurance systems concerned with food handling. However, the actual practical automation of a quality control assistance system in warehouses, as a research field, has not yet been much explored. An investigation into the adoption of a decision support system (DSS) for automating the process may help to improve the situation. The purpose of this paper is to outline and illustrate a decision support approach to automate the existing human based decision making practice for determining the appropriate quality assurance operations for food inventory management. An integrative prototype system, namely, the Intelligent Food Quality Assurance System (IFQAS), has been developed, not only for facilitating the selection of the most appropriate quality control operations, but also for suggesting the best storage environment for the goods after the quality
5252
S.I. Lao et al. / Expert Systems with Applications 39 (2012) 5251–5261
has been checked. In the system, a Case-based Reasoning (CBR) engine which solves new cases by reusing the previous handling experience in decision making is proposed to replace the manual decision making patterns in deciding the quality assurance operations. Because of the special nature of the food items it is not possible for humans to judge accurately the condition of the food inventory or express the conditions in precise numerical values, so often vague, fuzzy logic techniques are applied in order to extract the critical quality assurance information in terms of fuzzy rules. Therefore, a Fuzzy rule-based Reasoning (FBR) engine is constructed for suggesting the appropriate storage environment. The rest of this paper is organized as follows. In Section 2, the current quality assurance method used in the food industry with the application of various kinds of technology is reviewed. In Section 3, the system architecture of an Intelligent Food Quality Assurance System (IFQAS) and the mechanism of its two modules are described and explained in detail. In Section 4, a case study for validating the feasibility of the adoption of IFQAS is presented. Section 5 contains a detailed discussion of the system’s performance. The final section, Section 6, concludes the paper. 2. Related works Deterioration in food quality may lead to a huge impact on the credibility of a food company (Shapton & Shapton, 1991) and products that have been subjected to a compromise of different conditions and sensory properties would affect customer’s choice and preference during purchasing (Perrot et al., 2004). Now, there is a movement afoot to change the current human-judgment based food quality control approach to a more scientific and systematic one. When the quality and conditions of food are solely interpreted by the human operators, errors occur easily (Perrot et al., 2006). Human based quality assurance or decision making can manage the operations, but are believed to be unimaginative and arduous and lead to inconsistent actions (Davidson, Ryks, & Chu, 2001). Therefore, solely relying on human observation and evaluation is not systematic or reliable enough. This indicates that implementing quality assurance or control measurement is a critical issue. Researchers have explored this issue, however, it was found that their emphasis was on the importance of implementing such assurance system (Blaha, 1999), investigating the relationship between factors, such as the consequence of poor quality, the quality culture and the effectiveness of the goals achieved and the activities which lead to its adoption, (Karipidis, Athanassiadis, Aggelopoulos, & Giompliakis, 2009), and the provision of an implementation plan (Asefa et al., 2011). Researchers rarely evaluated the decision making process used for selecting the appropriate quality control operation procedures for particular types of food. In addition to this, researchers have adopted different DSS when undertaking quality control operations. Deslandres and Pierreval (1997) have designed a knowledge advisory system for quality applications; a quality advisory model is built for structuring the knowledge required for quality problem solving. Carpenter and Maropoulos (2000) have developed a system called OPTIMUM for controlling the tool machining process. It makes use of a combination of mathematical modeling and rule-based statistical methods for decision making. Shaffer and Brodahl (1998) have proposed a rule-based management system for offering management solutions for maintaining better farming conditions. Their predefined rules are stored in a database for decision making. However, those systems lack the self-learning capability of being able to procure human being’s knowledge. In handling food related operations, the decision making process involves numerous adjustments, therefore, self-learning and improving are important. As it has this special skill it is believed that Case-based Reasoning (CBR) is probably the right choice. CBR is one of the most popular
techniques for solving problems in the selection of operations. Peng, Chen, Wu, Xin, and Jing (2011) have developed a virtual reality based integrated system, applying CBR for matching the most similar machining fixture design case from a case-based database. It helps assign the production operations in the designing of parts. Kalapanidas and Avouris (2001) have proposed a NEMO prototype for improving the quality of air in operating conditions. It combines heuristic and statistical techniques based on the CBR approach for suggesting solutions. Chow, Choy, Lee, and Lau (2006) have applies CBR for selecting the most suitable resource usage package for handling warehouse operations. It found that using CBR engine for suggesting the operations is time saving and cost effective. Xia and Rao (1999) adopted CBR technology in an operation support system. The implementation result pointed out that CBR helps achieve more consistent and accurate operations. It found that there has been a trend to adopt learning techniques which use cases or instances directly, such as in Case-based Reasoning (CBR) (Kim, Im, & Park, 2010). In general, it is believed that CBR is particularly suitable for solving problems of domains that are experience-rich but knowledge poor (Chi, Chen, & Kiang, 1993). Nevertheless, the CBR techniques may only be capable of handling the decision making process at the operations selection level, but it may not be applicable for solving problems in food handling such as monitoring storage conditions. To reach a highquality level and identify errors in operations, attention should be paid to all of the characteristics of the sub-processes (Guillaume & Charnomordic, 2001). Probably, a system which can provide an all-round quality assurance support down to the parameter level is required. With reference to the published studies, Pacella and Semeraro (2005) applied Artificial Neural Networks (ANN) to the management of quality control in manufacturing processes. It recalled the learning patterns from the incomplete representations in order to monitor the quality characteristics. The system helps detect, classify and predict any unnatural changes that occur in the manufacturing process. Bezerra et al. (2007) have used ANN for data classification and pattern recognition for predicting material behavior for quality prediction. Jung and Yum (2011) have proposed an ANN-based approach for mapping the relationship between the characteristics, design and signals for taguchi parameter design. However, ANN need their inputs to be expressed in numeric terms for the decision making to process (Metaxiotis, Ergazakis, Samouilidis, & Psarras, 2003), the relationship between the status and the condition of the food items as the storage environment evaluation is complex and difficult to describe through modeling. Instead of using modeling, tackling this relationship using fuzzy logic approach is a probable solution. Fuzzy logic provides a methodology that managing blurry attributes and allow the use of data and information from those who possess expert knowledge. It has become an increasing important approach for tackling food problems and handles human reasoning in linguistic terms (Guillaume & Charnomordic, 2001). It is found that fuzzy logic techniques have been widely used in managing food problems. Researchers (Ioannou, Perrot, Curt, Mauris, & Trystram, 2004; Lababidi & Baker, 2003; Perrot et al., 2004) have adopted the fuzzy logic in assisting the quality control of food items during the production process. It is found that the concept and terms describing the quality of food items, and the justifications in the human mind are in an area of uncertainty and vagueness. Therefore, it is difficult to adjust and make decisions based on the measurement results in terms of numbering and crisp values. Fuzzy logic is considered as a tool that is suitable for dealing with fuzzy relationships, criteria and phenomena (Amelia, Wahab, & Hassan, 2009; Lin & Hsieh, 2004). Applying fuzzy terms in decision making, helps capture the reasoning process or the hidden uncertainty of operations (Jiang & Chen, 2005). Hence, the development of fuzzy sets helps build the linkage between words and numbers.
S.I. Lao et al. / Expert Systems with Applications 39 (2012) 5251–5261
5253
Fig. 1. Operation flow of the Intelligent Food Quality Assurance System.
Despite the wide application of fuzzy logic in food production monitoring, the technique has yet to be applied in the area of food safety assurance. Therefore, fuzzy logic is proposed in this paper for suggesting the storage conditions in warehouses. After examining the special needs and characteristics of food, this study attempts to propose a DSS using CBR and fuzzy logic to determine the appropriate quality assurance operations for food inventory management.
Hence, other than operation guidelines, a set of suggested storage environmental conditions is generated with the help of the fuzzy inference module by evaluating the food conditions. These suggestions further provide monitoring of the conditions down to the parameter level. In Fig. 2, the system architecture is shown. The Intelligent Food Quality Assurance System (IFQAS) consists of three tiers: the system database module, CBR module, and FBR module. The detailed operation mechanism of the system is described as follows:
3. The Intelligent Food Quality Assurance System (IFQAS) In this section, the Intelligent Food Quality Assurance System (IFQAS) is presented that provides knowledge based decision support for food management. The proposed IFQAS controls the food quality assurance operations in a warehouse by gathering different food attributes within the receiving and storage process, and converting the data into knowledge in terms of case and fuzzy rules. Quality assurance operations and the optimal storage environment are suggested based on the knowledge stored in the knowledge domain. As shown in Fig. 1, order specifications of the inbound inventory which are gathered from the workplace and the company database are collected. With the assistance of the CBR engine, suggested operation guidelines for quality assurance are generated.
3.1. Database module In the database module, different inventory parameters are stored. These data are grouped into two types: Order Specifications and Food Conditions. Order Specifications consist of physical information that is specified in the order invoices and on packaging, such as location of origin, quantity, storage type, and commodity name. This information is required for visualizing the categories and requirements of the foods. Information regarding Food Conditions is related to the environmental conditions and quality of the various foods. These data include temperature, color, elasticity and product size.
Fig. 2. System architecture of Intelligent Food Quality Assurance System.
5254
S.I. Lao et al. / Expert Systems with Applications 39 (2012) 5251–5261
3.2. CBR module In this tier, a CBR module is constructed to provide suggestions for the quality assurance operations. The CBR engine retrieves information about the inventories and about the existing quality standards of the company. The data act as the attributes for the retrieval of historical cases. Cases with the highest similarity value to the new case would be adopted. Modification may take place to enhance the case suitability. There are four phases in the CBR module: (i) Retrieve, (ii) Reuse, (iii) Revise, and (iv) Retain.
Fig. 3. IFQAS implementation steps.
(i) Retrieve Previous records are treated as cases and are divided into problem and solution parts. The problem part includes relevant case features, and the solution part mentions handling conclusion and comments. When food arrives, the CBR engine would activates the system to browse for the required information for the generation of key attributes. Using the list of generated key attributes, the CBR engine searches for potentially useful cases in the case library. Inductive indexing is applied for clustering cases with various similarities to the case being handled. Hence, a group of potentially useful cases in the resulting cluster with the tree structured searching, are generated and presented for case reuse. (ii) Reuse With the potential cases list generated, similarity values of the potential cases to the newly cases are calculated based on Eq. (1). The total similarity value is calculated base on the summing up of the similarity value of all attributes with weightings. With reference to all the total similarity values, cases are ranked in descending order:
Pn
I R i¼1 wi simðfi ; fi Þ Pn i¼1 wi
on the food safety during storage. Based on the food conditions, an appropriate storage environment is suggested, so as to help maintain the food quality. The controlling parameter is storage temperature. The fuzzy inference engine is composed of a rule-based reasoning engine that using fuzzy logic for solution suggestions. Generally, there are three components in the fuzzy inference: a fuzzifier for translating the crisp inputs to fuzzy values; an inference engine for applying a fuzzy reasoning mechanism to obtain a fuzzy output and a defuzzifier for translating a fuzzy output into a crisp value. Besides these, there is a knowledge base which contains fuzzy rules in terms of a rule-base and membership functions as a database. The characteristics of the fuzzy set are determined by membership functions. The present conditions of the food inventories need to be gathered so that they can act as the measurement data of the engine. Those data are then transformed into the right format, and act as the input data set of the fuzzy system. Knowledge from the workers and the implicit recording of human decisions and process are extracted and then analyzed. This knowledge is then turned into IF-THEN rules and stored into the rule set database for decision making. Those rules are part of the fuzzy system.
ð1Þ
where wi is the weight of attribute I and fiI ; fiR are the value of fi in the input. The case ranked with the highest similarity value is treated as the most significant case for action suggestion. The final operations suggestion would mainly consider the most highly ranked case. Cases with comparatively lower total similarity value would only serve as a reference in the revision stage.
3.3.1. Fuzzification Generally, there are four membership functions used in the engine. Fuzzy sets are defined by membership functions and each of the elements is designated as belonging to a degree of membership (between 0 and 1). A control framework is built which makes decisions based on different dimensions of the product quality, such as temperature, color, elasticity and product size. Fuzzy set temperature (T):
(iii) Revise When a historical case is suggested by the CBR engine, modifications, like editing and combining, are carried out if necessary. This step ensures the final case suggestion fits the actual situation and increases the adoptability of the solution. (iv) Retain The generated revised report and the new case content are then sent to the case library for storage. It acts as a reference and case asset for future use. After the CBR engine has been used for a period of time, the case database maturity increases. 3.3. Fuzzy rule-based module In this tier, a fuzzy inference module namely, Fuzzy rule-based module (FRB), is constructed for further processing of the CBR case results. With the generated CBR’s case suggestion, those foods that require temperature control monitoring are identified. In order to further ensure a holistic quality control to be performed, the fuzzy inference module provides expert advice and qualitative prediction
Fig. 4. Operation platform of IFQAS.
S.I. Lao et al. / Expert Systems with Applications 39 (2012) 5251–5261
5255
Fig. 5. System input of the relationship and importance among attributes.
Fig. 7. Quality Assurance Guidelines suggested from the CBR module. Fig. 6. Quality Assurance Guidelines cases suggestion.
T¼
n X
lT ðti Þ
ij1
ti
T, the whole data set; t, element of subset T. Fuzzy set Color (C):
C¼
n X
lC ðci Þ
ij1
ci
C, the whole data set; c, element of subset C. Fuzzy set Elasticity (E):
E¼
n X
lE ðei Þ
ij1
ei
E, the whole data set; e, element of subset E. Fuzzy set Size (S):
S¼
n X
lS ðsi Þ
ij1
si
S, the whole data set; s, element of subset S. In each of the membership functions, different membership values are defined. Those values are named the universe of discourse. It is divided into several regions for prediction of the food conditions. This can be exemplified by using temperature:
Temperature ðTÞ : fH; RH; M; RL; Lg where are high (H), relatively high (RH), medium (M), relatively low (RL), low (L). With the input data, the corresponding membership values of different attributes are calculated. 3.3.2. Inference engine The quality perception mechanism and the relationship between fuzzy input and fuzzy output variables are presented in
5256
S.I. Lao et al. / Expert Systems with Applications 39 (2012) 5251–5261
Fig. 8. The architecture of FBR.
the form of simple and intuitive ‘‘IF-THEN’’ fuzzy rules. The fuzzy rules were developed based on the results of interviewing quality experts, evaluating the industrial data and referring to the literature. The set of rules of the fuzzy relation are formulated, for example: Rule 1: IF A is high AND B is low THEN C is good. Rule 2: IF A is very low AND B is medium THEN C is bad. 3.3.3. Defuzzification To defuzzify the output fuzzy terms, the center of area (COA) is adopted to convert the fuzzy output back to a crisp value. The generating equation is
Pn xi Ai X ¼ Pi¼1 i¼1 Ai
where x is the center of gravity representing the probability of the case occurring; Ai denotes the area for each individual result. By calculating the COA, the fuzzy output is converted into crisp values with reference to the output membership function.
4. Case study In order to validate the effectiveness of the system, a pilot test was undertaken in one of the Hong Kong based Logistics companies. This company mainly handles food related inventories. Salmon fish is one of the core foods that this company handles. When salmon is received, it undergoes quality checking, customerized value added activities, and storing and then the salmon is delivered to the retail shops. Different quality checking and storage procedures are required for salmon which originates from different
Fig. 9. Input membership function (Color Score).
S.I. Lao et al. / Expert Systems with Applications 39 (2012) 5251–5261
5257
Fig. 10. Output membership function (Storage Temperature).
places, and which has been kept under different conditions and which is of different categories. Although quality assurance is required as part of the receiving process, the increase in value added activities increases the variety of the control points required and the difficulties in managing the operations. All of the decision making processes rely on the experience and knowledge of workers. There are no standard rules or regulations. Because of the numerous problems entailed in managing all the receiving operations, a trial run on IFQAS was undertaken in the company in August 2010 for a month. There are four implementation steps in total as illustrated in Fig. 3. 4.1. Data retrieval in the Warehouse Management System (WMS) Initially, the company adopted WMS for managing the warehouse inventory assignment functions. When goods are inbound
into the warehouse, data and information such as the quantity, SKU, the SKU dimensions, commodity name and the storage type are input into the WMS. In order to extract information for the CBR engine, an operation platform is built for data extraction. The operators key in the item number and order number to the system, click the ‘‘NEXT’’ button, and the screen will show the details of the order. This information is important for the evaluation of the inventory quality assurance requirements of the CBR module (Fig. 4). 4.2. Case construction of the CBR engine in IFQAS While constructing the CBR engine, two types of information are required: (i) The specific attributes for decision making. The usual practise of the company when managing the quality assurance operations is to rely on human justification based on the knowledge
Fig. 11. Fuzzy rule set.
5258
S.I. Lao et al. / Expert Systems with Applications 39 (2012) 5251–5261
Table 1 Matlab data code of the input variable ‘‘Color Score’’. [Input1] Name = ‘Color_Score’ Range = [20 34] NumMFs = 5 MF1 = ‘Poor’: ‘trapmf’, [19 20 21 24] MF2 = ‘Slightly_Poor’: ‘trimf’, [21 24 27] MF3 = ‘Medium’: ‘trimf’, [24 27 30] MF4 = ‘Good’: ‘trimf’, [27 30 33] MF5 = ‘Very_Good’: ‘trapmf’, [30 33 34 35]
of the staff about food management and the requirements of different food regulations such as Hazard Analysis and Critical Control Points (HACCP). Therefore, before turning the humanbased decision making process into a case-based decision DSS, the knowledge and the experience of the workers in the quality assurance department of the company is extracted and converted into cases. Several interviews are conducted for converting the human tacit knowledge into cases. Besides, all the previous salmon handling records of the company are studied, restructured and presented as historical cases. Those cases are then stored in the case library with attributes categorized into four levels: Level (1) Storage Types; Level (2) Customer Requirements; Level (3) Operation Specifications; and Level (4) Food Types. (ii) The relationship and importance of the case attributes. By defining the relationship and the importance of attributes, different weightings (wi) can be assigned. Hence, the similarity value (fi) among attributes can be defined. The entity’s importance for decision making and relationship to various attributes are then input into the CBR engine as shown in Fig. 5. With the case information provided by the WMS received by the CBR engine, a number of potentially useful cases are suggested. By calculating the total similarity value (TSV) through the CBR engine with reference to Eq. (1), which is shown in Section 3.2, cases are ranked. Cases with the highest TSV are selected as the most preferred cases for case adoption. As shown in Fig. 6, ‘‘Salmon 0325’’ with a TSV of 95 is selected.
Table 2 Sample rule sets of the system. Rule: 1 IF
(CS) (E) (T) (S)
Color Score of salmon fillet is very good AND Elasticity of meat is elastic AND Temperature of salmon is very low AND Size of salmon is very small
(ST)
Storage Temperature is no change
(CS) (E) (T) (S)
Color Score of salmon fillet is very good AND Elasticity of meat is elastic AND Temperature of salmon is low AND Size of salmon is small
(ST)
Storage Temperature is no change
(CS) (E) (T) (S)
Color Score of salmon fillet is good AND Elasticity of meat is elastic AND Temperature of salmon is very high AND Size of salmon is small
(ST)
Storage Temperature is slightly decreased
(CS) (E) (T) (S)
Color Score of salmon fillet is good AND Elasticity of meat is medium AND Temperature of salmon is medium AND Size of salmon is very small
(ST)
Storage Temperature is slightly decreased
(CS) (E) (T) (S)
Color Score of salmon fillet is slightly poor AND Elasticity of meat is medium AND Temperature of salmon is medium AND Size of salmon is medium
(ST)
Storage Temperature is decreased
(CS) (E) (T) (S)
Color Score of salmon fillet is good AND Elasticity of meat is medium AND Temperature of salmon is high AND Size of salmon is medium
(ST)
Storage Temperature is decreased
(CS) (E) (T) (S)
Color Score of salmon fillet is poor AND Elasticity of meat is inelastic AND Temperature of salmon is very high AND Size of salmon is very large
(ST)
Storage Temperature is heavily decreased
(CS) (E) (T) (S)
Color Score of salmon fillet is very good AND Elasticity of meat is very inelastic AND Temperature of salmon is medium AND Size of salmon is small
(ST)
Storage Temperature is heavily decreased
THEN Rule: 2 IF
THEN Rule: 3 IF
THEN Rule: 4 IF
THEN Rule: 5 IF
THEN Rule: 6 IF
THEN Rule: 7 IF
4.3. Fuzzy inference engine for food safety forecasting
THEN
With the Quality Assurance Guidelines suggested from the CBR module as shown in Fig. 7, the operators are provided with information and guidelines as to what quality assurance operations should be performed and how to perform them upon receipt of the salmon. As salmon belongs to the category that required temperature controls during storage in the warehouse, further operations in storage temperature suggestion is performed. Sensory evaluation is an important method for assessing the freshness and quality of fish. However, fuzzy terms are involved during the analyzing process. Therefore, a fuzzy inference engine is developed for evaluating the initial conditions of the salmon upon receiving it. Based on the conditions of the salmon, optimal storage conditions are suggested. In the past, the quality of the salmon was evaluated by human experts, manually, based on personal observations. Those experts were trained to analyze the condition of the salmon so as to decide the most appropriate storage environment conditions. Basically, the salmon fishes are stored at 4 °C. If the salmon was found to have deteriorated, it will be rejected. However, if the salmon has only slightly degraded, the storage temperature will be adjusted. The aim of making this adjustment is to help lengthen the product
Rule: 8 IF
THEN
life of the salmon by providing an appropriate storage environment. Interviews and evaluation of the company’s historical data was performed in order to convert this tacit and unstructured knowledge and information into rules and facts. By recording the knowledge of handling those processes, hence deducing the hidden methods of food handling, a number of rule sets were formulated. A structured fuzzy knowledge system was constructed using the Fuzzy Logic Toolbox of Matlab (Version 7.6). The structure of the FBR module for storage condition suggestion is shown in Fig. 8. Color, temperature, elasticity and size of salmon products, found to be important quality parameters (Anderson, 2000; Gallart-Jornet, Rustad, Barat, Fito, & Escriche, 2007), are used for
S.I. Lao et al. / Expert Systems with Applications 39 (2012) 5251–5261
5259
Fig. 12. List of fuzzy rules and the linguistic suggestions for storage temperature.
deciding the storage temperature. Those parameters are input into the system for evaluation. 4.3.1. Input membership functions All the four membership functions are defined and input into the system. Fig. 9 shows the membership functions plotted for one of the input variables, ‘‘Color Score’’, as an example. The input variable ‘‘Color Score’’ is a score for reflecting the freshness of salmon with reference to an international color ruler namely SalmoFan. The color of the salmon is one of the most important parameters for accessing its quality. A five point scale is used: Poor (P), Slightly Poor (SP), Medium (M), Good (G), Very Good (VG).
Color ScoreðCSÞ :fP; SP; M; G; VGg Table 1 shows the data code for indicating the difference of different fuzzy regions. Those coding are input into the system. 4.3.2. Output membership function The system has one output linguistic variable; ‘‘Storage Temperature’’ (ST). It has three terms for instructing different corrective actions as shown in Fig. 10. Based on the expert knowledge, the terms are defined and input into the system for indicating the relationship in the membership function. The output membership function ‘‘Storage Temperature’’ is divided into four regions: No change (N), Slightly Decrease (SD), Decrease (D), Heavily Decrease
Storage TemperatureðSTÞ : fN; SD; D; HDg
4.3.3. Fuzzy rules In order to analyze the conditions and quality of the salmon, different rules are set up based on the knowledge of salmon storage, as shown in Fig. 11. Table 2 displaces the sample rule set of the system. By using the rule viewer to input details of the parameters using the graphical expressing diagram as show in Fig. 12, the system automatically suggests the optimal storage condition for salmon storage in the warehouse. In this case, a salmon with a ‘‘Color Score’’ of 28.4, ‘‘Elasticity’’ of 0.762 s, receiving ‘‘Temperature’’ of 1.2 °C and salmon ‘‘Size’’ of 3.86 kg is input for analysis. The system calculates the compatibility of each rule, and then infers a conclusion for each rule. The inferred conclusions from all rules with the same linguistic value are aggregated and combined into a single fuzzy set. Hence, the systems truncating the fuzzy output membership function into numerical value by defuzzification. Finally, a storage temperature of 1.25 °C is suggested. 5. Results and discussion After the pilot test of IFQAS, the results showing the effectiveness of the system are analyzed. The performance of both the traditional, human based approach and the DSS assistance approach with IFQAS are compared in the following section. Two aspects of performance are examined; they are (1) improvement in quality control and (2) improvement in operation efficiency.
5260
S.I. Lao et al. / Expert Systems with Applications 39 (2012) 5251–5261
5.1. Quality control improvement
6. Conclusions
As shown in the case study section, operation guidelines are suggested for all inbound inventories and the storage conditions are further suggested according to the type of inventory that is under temperature control. The operation guidelines are userfriendly which assists the operators to master the food quality assurance process in accordance with the customer requirements, value-added activities and the inventories specifications. By analyzing the data with the CBR engine, the operations that need to be performed for the purpose of quality assurance are shortlisted. The FBR engine then follows the criteria to further suggest the judgment solution for maintaining the inventory quality. Operators can perform specific actions according to the determined guidelines and adjustments. Before this system was implemented, a chaotic manual decision making approach was adopted, which led to errors and inconsistent decisions. IFQAS help to provide holistic quality control of the receiving operations down to parameter levels which improved the inventory defective rate and the customer satisfaction level, as shown in Table 3.
In a warehouse, quality control of inbound inventory is critical. Receiving operations act as the first gate for quality assurance especially with food inventory. Burdensome procedures and regulations make it difficult for operators to carry out their duties efficiently. Hence, the traditional methods mainly considered the quality control activities at the operations level, they rarely focused on improvement at the parameter level. This paper provides a holistic monitoring approach from the receiving docks to the inventory storage stage. It demonstrated a decision support approach to automate the human based decision making process in the determination of quality assurance operations and storage conditions suggestion of inventory at the parameter level. With the support of the decision support technologies like CBR and FBR, a system, namely IFQAS, is designed to assist and automate the safety control decision making process. The aim of the system is to provide consistent and systematic quality assurance, with improvement in customer satisfaction level and in the defective rate. Despite the contributions of the proposed system in the food quality control perspective, the system is not without constraints. The tacit knowledge and operation practices of human experts are computerized for the preservation of past cases and to help in the construction of rules. In order to collect information for those formations, investigation or analytical tasks need to be executed. Such requirements may not be easy for small or immature companies to fulfill, which increases the difficulty of successful adoption. Thus, the generalization of the storage condition suggestions to other warehouse with different food inventories may be limited, as different parameters may need to be analyzed when some special food items are involved. Further investigation on the appropriateness of integrating the Radio Frequency Identification (RFID) technique for automating the process is suggested. RFID can help improve the visibility of the location and status of the inventory and this can further develop a holistic monitoring of the inventory.
5.2. Operation efficiency improvement The human-based decision making process involved manual analyzing, normally manual review on the inventory specifications and relevant quality guidelines, regulations and documents. The manual evaluation process usually required over 15 min, an even longer time frame was required if a new SKU is received. The IFQAS makes use of the data in the WMS by transferring them to the DSS, which reduces the data gathering time. Hence, the system supported decision making process can generate the operation guidelines and storage conditions automatically in a shorter timeframe, as shown in Table 4. As food inventory needs a short timeframe and fast supply chain for handling, the decrease in operation time helps improve the performance.
Table 3 Improvement in customer satisfaction and quality.
Table 4 Time reduction in operation handling.
S.I. Lao et al. / Expert Systems with Applications 39 (2012) 5251–5261
Acknowledgements The authors wish to thank the Research Office of the Hong Kong Polytechnic University for supporting the project (Project Code: RPBK). References Amelia, L., Wahab, D. A., & Hassan, A. (2009). Modeling of palm oil production using fuzzy expert system. Expert Systems with Applications, 36, 8735–8749. Anderson, S. (2000). Salmon color and the consumer. International Institute of Fisheries Economics and Trade, 1–3. Asefa, D. T., Kure, C. F., Gjerde, R. O., Langsrud, S., Omer, M. K., Nesbajjen, T., et al. (2011). A HACCP plan for mycotoxigenic hazards associated with Dry-cured meat production processes. Food Control, 22, 831–837. Bezerra, E. M., Ancelotti, A. C., Pardini, L. C., Rocco, J. A. F. F., Iha, K., & Ribeiro, CH. C. (2007). Artificial neural networks applied to epoxy composites reinforced with carbon and E-glass fibers: Analysis of the shear mechanical properties. Materials Science and Engineering, 464, 177–185. Blaha, Th. (1999). Epidemiology and quality assurance application to food safety. Preventive Veterinary Medicine, 39, 81–92. Carpenter, I. D., & Maropoulos, P. G. (2000). A flexible tool selection decision support system for milling operations. Journal of Materials Processing Technology, 107, 143–152. Chi, R. T., Chen, M., & Kiang, M. Y. (1993). Generalized case-based reasoning system for portfolio management. Expert Systems with Applications, 6, 67–76. Chow, H. K. H., Choy, K. L., Lee, W. B., & Lau, K. C. (2006). Design of a RFID case-based resource management system for warehouse operations. Expert Systems with Applications, 30, 561–576. Davidson, V. J., Ryks, J., & Chu, T. (2001). Fuzzy models to predict consumer ratings for biscuits based on digital image features. IEEE Transactions on Fuzzy Systems, 9(1), 62–67. Deslandres, V., & Pierreval, H. (1997). Knowledge acquisition issues in the design of decision support systems in quality control. European Journal of Operational Research, 103, 296–311. Gallart-Jornet, L., Rustad, T., Barat, J. M., Fito, P., & Escriche, I. (2007). Effect of superchilled storage on the freshness and salting behavior of Atlantic salmon (Salmo salar) fillets. Food Chemistry, 103, 1268–1281. Getinet, H., Seyoum, T., & Woldetsadik, K. (2008). The effect of cultivar, maturity stage and storage environment on quality of tomatoes. Journal of Food Engineering, 87, 467–478. Guillaume, S., & Charnomordic, B. (2001). Knowledge discovery for control purposes in food industry databases. Fuzzy Sets and Systems, 122, 487–497. Ioannou, I., Perrot, N., Curt, C., Mauris, G., & Trystram, G. (2004). Development of a control system using the fuzzy set theory applied to a browning process—A fuzzy symbolic approach for the measurement of product browning: Development of a diagnosis model—Part I. Journal of Food Engineering, 64(4), 497–506. Jiang, Q., & Chen, C. H. (2005). A multi-dimensional fuzzy decision support strategy. Decision Support Systems, 38, 591–598.
5261
Jung, J. R., & Yum, B. J. (2011). Artificial neural network based approach for dynamic parameter design. Expert Systems with Applications, 38, 504– 510. Kalapanidas, E., & Avouris, N. (2001). Short-term air quality prediction using a casebased classifier. Environmental Modeling & Software, 16, 263–272. Karipidis, P., Athanassiadis, K., Aggelopoulos, S., & Giompliakis, E. (2009). Factors affecting the adoption of quality assurance systems in small food enterprises. Food Control, 20, 93–98. Kim, H. K., Im, K. H., & Park, S. C. (2010). DSS for computer security incident response applying CBR and collaborative response. Expert Systems with Applications, 37, 852–870. Lababidi, H. M. S., & Baker, C. G. J. (2003). Web-based expert system for food dryer selection. Computers and Chemical Engineering, 27, 997–1009. Lin, C., & Hsieh, P. J. (2004). A fuzzy decision support system for strategic portfolio management. Decision Support Systems, 38, 383–398. Metaxiotis, K., Ergazakis, K., Samouilidis, E., & Psarras, J. (2003). Decision support through knowledge management: The role of the artificial intelligence. Information Management & Computer Security, 11(5), 216–221. Nilsson, L., Johnson, M. D., & Gustafsson, A. (2001). The impact of quality practices on customer satisfaction and business results: Product versus service organizations. Journal of Quality Management, 6, 5–27. Orriss, G. D., & Whitehead, A. J. (2000). Hazard analysis and critical control point (HACCP) as a part of an overall quality assurance system in international food trade. Food Control, 11, 345–351. Pacella, M., & Semeraro, Q. (2005). Understanding ART-based neural algorithms as statistical tools for manufacturing process quality control. Engineering Applications of Artificial Intelligence, 18, 645–662. Peng, G., Chen, G., Wu, C., Xin, H., & Jing, Y. (2011). Applying RBR and CBR to develop a VR based integrated system for machining fixture design. Expert System with Applications, 38, 26–38. Perrot, N., Agioux, L., Ioannou, I., Mauris, G., Corrieu, G., & Trystram, G. (2004). Decision support system design using the operator skill to control cheese ripening – Application of the fuzzy symbolic approach. Journal of Food Engineering, 64, 321–333. Perrot, N., Ioannou, I., Allais, I., Curt, C., Hossenlopp, J., & Trystram, G. (2006). Fuzzy concepts applied to food product quality control: A review. Fuzzy Sets and System, 157, 1145–1154. Shaffer, M. J., & Brodahl, M. K. (1998). Rule-based management for simulation in agricultural decision support systems. Computers and Electronics in Agriculture, 21, 135–152. Shapton, D. A., & Shapton, N. F. (1991). Principles and practices for the safe processing of foods. Oxford: Butterworth-Heinemann. Xia, Q., & Rao, M. (1999). Dynamic case-based reasoning for process operation support systems. Engineering Applications of Artificial Intelligence, 12, 343– 361. Yan, Z., Sousa-Gallagher, M. J., & Oliveira, F. A. R. (2008). Identification of critical quality parameters and optimal environment conditions of intermediate moisture content banana during storage. Journal of Food Engineering, 85, 163–172. Youngdahl, W. E., & Kellogg, D. L. (1997). The relationship between service customers’ quality assurance behaviors, satisfaction, and effort: A cost of quality perspective. Journal of Operations Management, 15, 19–32.