Knowledge Discovery Approach for Automated Process Planning

Available online at www.sciencedirect.com

ScienceDirect Procedia CIRP 63 (2017) 539 – 544

The 50th CIRP Conference on Manufacturing Systems

Knowledge discovery approach for automated process planning Guenther Schuh, Jan-Philipp Prote, Melanie Luckert, Philipp Hünnekes* Laboratory for Machine Tools and Production Engineering (WZL), RWTH Aachen University, Steinbachstraße 19, D-52074 Aachen, Germany * Corresponding author. Tel.: +49 241 80 20396; fax: +49 241 80 22293. E-mail address: [email protected]

Abstract Manufacturing companies in industrialized countries are facing the challenge of achieving shorter times-to-market for their products while also coping with higher and more frequent initial planning efforts for customer specific products. Automated process planning is suited to dissolve this conflict by reducing manual planning efforts and enhancing planning productivity. However, existing computer-aided process planning (CAPP) approaches primarily shift planning efforts towards establishing and updating deterministic rules for planning algorithms manually. This paper shows the potential of using feedback data from Industrie 4.0 production systems as well as design features in a statistical approach to automatically determine initial process sheet information for new products. Feedback data from the manufacturing system is used as a digital representation of the production process. Interdependencies of component characteristics and production processes can be statistically identified via a knowledge discovery in databases (KDD) approach. These interdependencies in turn can be used to automatically deduce rules for CAPP planning algorithms. The presented integrated approach also includes further increasing the level of accuracy and comprehensiveness of the initial process sheet information, as well as updating the planning rules and assumptions following a control loop model. Necessary input and output parameters of the approach are being described, as well as the approach itself, including several steps to systematically incorporate the implications of component characteristic interdependencies on the necessary process steps. Finally, the approach and its potentials are illustrated using a set of real data from a manufacturing company. © 2017 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license © 2017 The Authors. Published by Elsevier B.V. (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peer-review under responsibility of the scientific committee of The 50th CIRP Conference on Manufacturing Systems. Peer-review under responsibility of the scientific committee of The 50th CIRP Conference on Manufacturing Systems Keywords: CAPP; automated process planning, product-process-interdependencies, KDD, data mining

1. Introduction Ensuring a competitive time-to-market for product development and time-to-customer for order fulfillment respectively are important requirements for manufacturing companies [1]. These challenges are especially valid for individual and small batch production [2]. Decreasing product lifecycles [3] as well as the development towards customer individual product variants [4,5] lead to an increase in planning efforts per unit produced and the necessity to rationalize planning activities [3]. In addition to planning efficiency, the use of process sheets as a crucial document for production control activities, poses quality requirements on the planning process that are often not met in reality. The use of incorrect enterprise resource planning (ERP) master data and disregard of product and/or process

changes can lead to the use of incorrect process sheets [6,7]. Implicit planning knowledge of domain experts furthermore can lead to non-standardized, inconsistent planning processes and results [8]. To address the illustrated challenges regarding planning efficiency and quality, the automation of process planning tasks by using computer-aided process planning (CAPP) has been researched [1] and commercial software is in widespread use. Externalizing planning knowledge to manually set up and configure CAPP systems to be used productively however, is time consuming and intricate [1,9]. In order to further improve process planning efficiency, it is necessary to automate the externalization of process planning knowledge. Therefore, this paper proposes an approach to automatically elicit process planning knowledge by statistically analyzing interdependencies of component characteristics and production

2212-8271 © 2017 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peer-review under responsibility of the scientific committee of The 50th CIRP Conference on Manufacturing Systems

doi:10.1016/j.procir.2017.03.092

540

Guenther Schuh et al. / Procedia CIRP 63 (2017) 539 – 544

processes. These interdependencies in turn can be used to automatically deduce rules for CAPP planning algorithms. To ensure an up-to-date knowledge base at all times, feedback data from recent production orders is used as a representation of the production processes. The paper is structured as follows: Section 2 discusses process planning, feature technology, as well as CAPP approaches and existing approaches for automated discovery of planning knowledge. Section 3 subsequently presents the approach for automated process planning. The presented approach is then illustrated using a set of feedback data from an individual and small series manufacturer in section 4. This paper concludes with a summary of its findings and an outlook upon further research in section 5.

that can be easily interpreted by CAD, as well as CAPP systems [10]. For this purpose, component information is usually represented by features. Features contain the geometric description of the component elements, as well as their spatial relations. The use of feature-technology and feature-based CAPP systems is widely accepted in scientific publications and used in commercial software [1,10]. Approaches for featurebased design of components are distinguished from approaches for retroactive feature-identification from CAD-models [12]. A mostly consistent use of feature-descriptions among systems and platforms is achieved by employing the featuredescriptions standardized in STEP-files (Standard for the Exchange of Product Model Data) [10,13]. 2.3. Automation approaches to process planning

2. State of the art This section describes the fundamentals of process planning as the basis for production scheduling and control activities. Publications discussing CAPP approaches as well as approaches to automatically identify and elicit process planning knowledge from existing databases are being reviewed and a research gap is identified. 2.1. Process planning In the product creation process, process planning constitutes the activity that links product design and manufacturing. Process scheduling as a subsequent activity is then responsible to take appropriate action in order to ensure the operations are being carried out as planned. As a short term planning activity, process planning determines feasible and economically viable manufacturing sequences considering the available resources. Macro level process planning, as used for production scheduling and control activities, incorporates the four planning problems blank selection, process sequence determination, resource allocation and standard time determination. Planning results are then documented as process sheets (also process plans) and constitute the pivotal document for the manufacturing activities [1,3]. Traditionally, manufacturing experts, based on expertise and experience [8], carry out process planning manually. Information asymmetry among planning experts, insufficient feedback from manufacturing, unchecked reuse of process sheets, as well as the complexity of the planning problem itself, often lead to insufficient planning results. Considering these circumstances, CAPP systems are increasingly in widespread use in order to improve planning quality and efficiency [10]. However, compared to other computer-aided technologies like computer-aided design (CAD), CAPP systems have lower market penetration [10,11]. Considerable implementation efforts due to system architecture and the integration of implicit planning knowledge, specific to the implementing company [1,3] can be reasons for this circumstance. 2.2. Feature-technology In order to apply CAPP technology, it is necessary to specify the component and its characteristics geometrically in a way

Numerous approaches to automate aspects of the planning process have been researched and published for decades (cf. [8,14]). Generative planning approaches, which enable the automated process planning for new components, have been at the center of this research, but are also the most difficult to implement [10]. Approaches frequently used to link companyspecific product and process information in the automated planning process are decision trees and tables, the use of fuzzy rules and of artificial neural networks (ANN). Decision trees and table-based approaches have been used to solve various kinds of planning problems. Using defined ifthen relations, process information can be linked to the component based on its features and their parameter values [1]. Sadaiah et al. propose the use of a decision table-based approach for the planning of prismatic components [15]. The focus of the approach is an economically sensible and technologically feasible grouping of features and therefore assigned processes into machining set-ups. Lee at al. also pursue the goal of grouping necessary processes into an economically advantageous sequence of machining set-ups [16]. The emphasis in this approach is put on integrating associated features into “composite features”, as well as determining the standard times and prioritization of the processes. Other exemplary decision table-based approaches also focus on the micro planning level and assign processing operations within machining set-ups [17,18]. Raman et al. allocate machine tools to component features based on their ability to manufacture the required geometries and shapes [12]. An approach for macro level process planning is proposed by Nonaka et al. [11]. Machining volume is decomposed into basic shapes, which in turn are assigned to suitable resources, with the objective to generate alternative routings in order to facilitate a uniform utilization of capacities. Fuzzy logic-based approaches allow for a generalization of unambiguous decision rules used in decision trees and tables. Value ranges of input parameters can be attributed to several classes of output variables at the same time when using fuzzy logic [19]. Fuzzy logic can be used to determine the suitability of existing process plans for a new production order and for machine tool and resource selection as well as for choose machining parameters as a function of material properties as well as feature size [20].

Guenther Schuh et al. / Procedia CIRP 63 (2017) 539 – 544

A combination of fuzzy rules and ANN is used in a comprehensive process planning approach by Amaitik and Kiliç [21]. Neural networks connect multidimensional functions for classification of input variables (neurons) in a variety of layers and thus are able to perform complex classification tasks [22,23]. Amaitik and Kiliç use ANN to select manufacturing processes, resources and machine tools on the macro planning level. With the objective to facilitate real time dynamic rerouting in case of unforeseen events on the shopfloor, ANN are used by Joo et al. [24]. As shown in this section, various automation approaches for process planning problems exist. These approaches presume the existence of a structured database with explicit process planning knowledge in the form of if-then relations, fuzzy rules or the design of training algorithms and data sets for ANNbased approaches. The necessary comprehensive determination of abstract planning decisions in order to establish these structured company-specific process planning databases generates high implementation efforts [1,9]. In order to enable process planning according to the latest information, efforts for continuous analysis and modification of these databases are necessary as well [15]. 2.4. Extracting process planning knowledge from databases As described in section 2.1, process planning is a planning activity relying greatly on personal expertise and experience of manufacturing experts. To empower the automated planning approaches introduced in section 2.3, various approaches to identify and extract planning knowledge for process planning tasks from feedback data and from databases exist. The analysis of feedback data from previous production orders to define standard times manually is a common approach [1]. However, also approaches for the automated analysis of feedback data and deduction of standard times have been proposed. Reinhart and Geiger perform a continuous statistical analysis of recorded processing times and criteria-based updating of standard times considering the deviations of processing and current standard times [25]. The approach focuses on updating existing process plans in real-time and the improvements in production scheduling hereby enabled. Monostori et al. [26] describe a similar approach of updating standard times planned by CAPP software, using statistical tools to compare recorded processing to current standard times. In addition, sequences of resources deviating from those planned for by the CAPP system are being automatically registered and brought to the attention of the production planner. That way, initial CAPP planning results are being improved within the first units produced. In order to deduce initial planning information that can be used for the generation of process plans for new components from databases, several approaches using the Knowledge Discovery in Databases (KDD) process have been proposed. Frequently addressed planning problems are the selection of suitable machining parameters, discovery of typical process routes and the identification of alternative routings in order to improve scheduling flexibility. Denkena et al [27] describe a KDD-based approach to identify suitable machining parameters for future machining operations. Data sets of

feedback data similar to the machining task to be planned are identified. The machining parameters of the selected data sets are then grouped by clustering algorithms and analyzed concerning their process capabilities. With the objective of reutilisation and standardization of planning operations, Liu et al. propose a KDD-based approach for discovery of typical process routes in existing process plans [9]. Machining processes and process sequences are analyzed with a hierarchical clustering algorithm in order to identify similar process sequences. The identified clusters with typical process routes can be searched for features machined and reused. Standard times and resource allocation are not being considered. Schmidt improves on scheduling flexibility after order release, by choosing and adopting from several possible process sequences based on information on the production system [10]. At the same time, clustering algorithms in a KDDbased approach identify machining parameters that have been successful under similar circumstances on prior production orders. KDD-based approaches present an opportunity to reduce the costs and efforts required for setting up and updating databases for process planning rules, which can be used by commercial CAPP systems. The presented approaches use exiting process plans as input for the analysis and thus risk manifesting outdated or incomplete process plans. Focusing on machining parameters and typical process routes, no KDD-based approaches exist, that enable a comprehensive deduction of process planning rules for all four planning problems on macro planning level. 3. Approach for automated process planning The basic concept of the presented comprehensive approach is to deduce process planning rules for CAPP systems automatically, by statistically identifying interdependencies between component features and recorded feedback data from historic production orders. The planning results obtainable from CAPP systems employing these rules are improved during the first production runs, by updating the process plans based on feedback data of the specific product variant. The proposed approach has five steps to enrich existing automation approaches for process planning systematically, as shown in figure 1.

Fig. 1: Comprehensive planning approach

541

542

Guenther Schuh et al. / Procedia CIRP 63 (2017) 539 – 544

The component features to be included in the statistical analysis are selected in step 1. The selected component features and the associated feedback data from previous production orders are transformed and pre-processed in step 2. Statistical interdependencies between the component features and process elements are identified in step 3, including data mining and interpretation. A deduction of planning rules based on the identified interdependencies constitutes step 4. Hereby a structured database with process planning rules is created, which can be used by CAPP systems to generate process plans based on component feature input. The updating of planning information based on a comparison of planned and actual process information is described in step 5. 3.1. Selection of component features The selection of component features to be included in the analysis is subject to contradicting requirements. In order to thoroughly determine the product-process-interdependencies, a maximum number of distinguished features is required. The application of statistical operations in step 3 requires the existence of data points in sufficient quantity. Especially for the intended analysis of feature combinations, a sufficient frequency of feature combination occurrence in the data set has to be ensured. Furthermore, the amount of features to be selected is limited because a large number of attributes for analysis affects the efficiency of data mining methods [10]. In practical applications, the range of features and their parameter values that are automatically identified by software modules are limited, however. Following an iterative approach, all distinguishable features are being used in the analysis at first. Features and parameters value distinctions, that do not significantly change planning results on macro level, are then stricken from further analysis.

material numbers. Information on the ability of the different resources to perform manufacturing processes is included in the resource information. Beyond the feature selection of step 1, filtering by product families can be included in the case of a heterogeneous product range and production processes in order to create planning modules. The approach for reducing data inconsistencies for production planning and control activities proposed by Schuh et al. [29] is used for preprocessing the recorded feedback data. For the subsequent automated application of data mining techniques, product and process information are combined and saved to a data warehouse, suited for this kind of analysis [30,31]. During this transformation, feature combinations of the different material numbers are linked to the historic resource sequences of the respective material numbers. Following the data mining stages of step 3 of the approach, the information output of the various stages is continuously saved to the data warehouse and linked to the other input parameters of the next stage. 3.3. Data mining and interpretation The data mining and interpretation activities of the KDD process applied in step 3 of the approach are structured in three stages, as shown in table 1. Table 1: Data mining stages Stage

Input parameters

Addressed planning problem

1

Feature sets, resource sequences

Resource allocation

2

Resource sequence, resource capabilities matrix

Process sequence determination

3

Feature and resource combinations, feature parameter values

Standard time determination

3.2. Preprocessing and transformation The second step of the proposed approach includes the further selection, pre-processing and transformation of the input data for the statistical analysis of step 3. These activities are part of the KDD process described by Fayyad et al. [28], as shown in figure 2. Data Mining Transformation

Preprocessing Filtering

Interpretation !

Data Target Data

Preprocessed Patterns data Transformed data

Knowledge

Fig. 2: KDD process [28]

Input data for the KDD process are the component features and parameter values selected in step 1 for the material numbers of the product range, feedback data from the production system and resource capabilities information. The feedback data includes information on the blanks used, sequence of resources used, processing and set-up times as well as lot sizes of historic production orders for the different

The activities are structured considering the interdependencies of the four planning problems and the already available information. Knowledge discovery attempts regarding standard times for machining particular features for example, have to be conducted for each eligible resource individually because of the varying performance characteristics of the resources. An association of individual component features to a process step or resource however, is often not directly apparent in conventional feedback data. For a given product family and production system, blank types are often predetermined with the blank size being a function of the size of the finished component. If it is analytically determined, that blank type and therefore process sequence are variable e.g. with lot size, a dedicated analysis to discover dependencies is advised. In order to obtain an eligible allocation of resources to a component characterized by its set of features, the first data mining stage explores the probabilities of engaging individual resources in the production process of the various feature sets. Algorithms using probabilistic classifiers, like divide-andconquer algorithms relying on information gains to construct classification trees by recursive partitioning, can be used for discovery of the product process interdependencies in the

Guenther Schuh et al. / Procedia CIRP 63 (2017) 539 – 544

various data mining stages [32]. The use of recursive partitioning also offers evidence-based information on what component characteristics (e.g. component dimensions, features) influence the production process to what extent to the design department. In order to avoid overfitting, minimum information gain requirements is used when cross-validation checks demonstrate an unfavorable ratio of parameters to observations. Having established eligible resource combinations for production of the various feature sets in stage 1, the most probable sequences for the resource combinations is then determined. Using the resource capabilities information, the feature sets and possible recordings of process information as input for analysis, the process sequences are then derived from the resource sequences in stage 2. Standard times are finally determined in stage 3 for individual feature to resource combinations, due to the unique specifications and capabilities of the resources, depending on the parameter values of the features.

4. Application The described approach has been applied to a data set of a recent case study at the Laboratory for Machine Tools and Production Engineering. The example provided is intended to illustrate parts of the approach, including data mining stage 1. Component features and order data of 160 historic production orders of piston variants have been analyzed using the data analysis software R (The R Project, freeware). The component features selected for analysis are weight, outside and inside diameters, length as well as number and position of drillings. The feature sets of the piston variants have been linked to the sequence of resources used in the historic production orders in preparation for analysis. In order to determine the decision tables for resource allocation, classification trees are calculated for each of the resources used in the data set. An exemplary classification tree for resource 543 is shown in figure 3.

3.4. Deduction of planning rules The interdependencies identified in the data mining stages have to be transformed into planning rules to be usable for CAPP planning algorithms. The information from classification trees and outputs of other classifying algorithms can easily be transformed, stored in and retrieved from interlinked decision tables. When pursuing a standardizationdriven approach, the statistically most likely output for a given set of input values is entered into the decision table. For flexibility-driven approaches with more complex CAPP algorithms, alternatively more than one possible output can be saved to the decision tables. 3.5. Updating of planning information After the initial establishment of planning rules for use in a CAPP system, the updating of planning information is a twofold process. First, deviations of the actual production processes from the corresponding process plans are being monitored continuously. Adjusted for infeasible data and outliers, process plans are updated to a moving average regarding resource allocation to process steps as well as standard times. This approach suggested in Monostori et al. [26] reduces the average process deviation from plan and thus creates process plans, which constitute a more reliable basis for scheduling activities. Updating the initial set of process planning rules by reexecuting steps 2 to 4 of the approach is not performed continuously, due to constraints regarding system resources. This second updating cycle is instead prompted by statistically significant process to plan deviations over a minimum number of orders regarding a given material number. The updating then is restricted to material numbers containing those features, that the material number prompting the updating possesses as well.

Fig. 3: Classification tree for resource 543

The classification tree in Fig. 3 shows, among other things, that components which include drillings on the face (node 1), being produced in a lot size > 16 (node 5) and are over 35.5 mm in length (node 7) always include machining on resource 543. The calculated parameter value intervals of the inner panel nodes and probabilities of the terminal nodes of the classification trees for all the included resources are then transferred into the decision tables for resource allocation. A classification tree is also generated regarding the probable number of process steps for a given feature set, in order to be able to allocate a finite number of resources to a component. Resources are then allocated to the components according to descending probabilities of utilization until the identified number of process steps is reached. It is possible for the classification trees to include nodes that demand decisions based on the previous allocation of other resources. Machining centres for example, might not be used for a milling operation if a lathe is previously allocated for a turning operation and vice versa. However, once a resource is

543

544

Guenther Schuh et al. / Procedia CIRP 63 (2017) 539 – 544

allocated to a component, this additional information is available for choosing the next most likely resource from the decision table. 5. Conclusion and further research In this paper, a comprehensive approach for the automatic deduction and subsequent update of process planning rules from recorded feedback data of historic production orders has been presented. The selection of component features for analysis and their preparation and linking to the feedback have been discussed. A sequence of data mining stages to address the four planning problems, as well as the deduction of planning rules from the information discovered were outlined and the first steps of the approach have been illustrated using real production data. Future research will be conducted especially regarding choosing and adjusting effective data mining approaches to address the remaining planning problems of step 3 of the approach. The inclusion of complementary information, like positively identifying processes on resources by searching the NC-programming for key words, might improve on planning quality and simplify the succession of data mining stages. The potential of the approach to provide an automated deduction of planning rules, while maintaining an adequate quality of initial planning information, has to be evaluated when the remaining steps of the approach have been implemented. 6. Acknowledgements The authors would like to thank the German Research Foundation DFG for the kind support within the Cluster of Excellence “Integrative Production Technology for High-Wage Countries”. References [1] Eversheim, W. Organisation in der Produktionstechnik. 3rd ed. Berlin: Springer, 2002. [2] Reuter, C.; Nuyken, T.; Schmitz, S.; Dany, S. Iterative Improvement of Process Planning Within Individual and Small Batch Production. In: Umeda, S., Nakano, M., Mizuyama, H., Hibino, N., Kiritsis, D., Cieminski, G. von, eds. Advances in Production Management Systems: Innovative Production Management Towards Sustainable Growth. Cham: Springer, 2015. p. 283–290. [3] Wiendahl, H.-P. Betriebsorganisation für Ingenieure, 8th ed. München: Hanser, 2014. [4] Gerritsen, B. Advances in mass customization and adaptive manufacturing. In: Horváth, I., Rusák, Z., eds. Tools and methods of competitive engineering, Proceedings of the Seventh international symposium on tools and methods of competitive engineering - TMCE 2008, April 21-25, Izmir, Turkey. Delft: University of Technology, 2008, p. 869–880. [5] ElMaraghy, H., Schuh, G., ElMaraghy, W., Piller, F., Schönsleben, P., Tseng, M., & Bernard, A. Product variety management. CIRP AnnalsManufacturing Technology 2013. 62(2): 629-652. [6] Kletti, J.; Schumacher, J.: Die perfekte Produktion. Berlin: Springer, 2011. [7] Jodlbauer, H.; Palmetshofer, K.; Reitner, S. Implizite Determinierung von Plan-Belegungszeiten. Wirtschaftsinformatik 2005. 47(2): 101–108. [8] Xu, X.; Wang, L.; Newman, S. Computer-aided process planning – A critical review of recent developments and future trends. International Journal of Computer Integrated Manufacturing 2011. 24(1): 1–31.

[9] Liu, S.; Zhang, Z.; Tian, X. A typical process route discovery method based on clustering analysis. The International Journal of Advanced Manufacturing Technology 2007. 35(1-2): 186–194. [10] Schmidt, J. Integrierte Arbeitsplanung und Fertigungssteuerung auf Basis von Zustandsinformationen. Denkena, B. ed. Garbsen: PZH, 2015. [11] Nonaka, Y.; ErdĘs, G.; Kis, T.; Kovács, A.; Monostori, L.; Nakano, T.; Váncza, J. Generating alternative process plans for complex parts. CIRP Annals - Manufacturing Technology 2013. 62(1): 453–458. [12] Raman, R.; Marefat, M. Integrated process planning using tool/process capabilities and heuristic search. Journal of Intelligent Manufacturing 2004. 15(2): 141–174. [13] ISO 10303-224 (15.08.2006): Industrial automation systems and integration - Product data representation and exchange - Part 224. [14] Yusof, Y.; Latif, K. Survey on computer-aided process planning. The International Journal of Advanced Manufacturing Technology 2014. 75(1-4): 77–89. [15] Sadaiah, M.; Yadav, D.; Mohanram, P.; Radhakrishnan, P. A Generative Computer-Aided Process Planning System for Prismatic Components. The International Journal of Advanced Manufacturing Technology 2002. 20(10): 709–719. [16] Lee, H.; Jhee, W.; Park, H.-S. Generative CAPP through projective feature recognition. Computers & Industrial Engineering 2007. 53(2): 241–246. [17] Pham, D.; Gologlu, C. A computer aided process planning system for concurrent engineering. Proceedings of the Institution of Mechanical Engineers, Part B: Journal of Engineering Manufacture 2001. 215(8): 1117–1131. [18] Sormaz, D.; Khoshnevis, B. Generation of alternative process plans in integrated manufacturing systems. Journal of Intelligent Manufacturing 2003. 14(6): 509–526. [19] Amaitik, S. Fuzzy Logic Models for Selection of Machining Parameters in CAPP Systems. International Journal of Computer and Information Technology 2013. 2(2): 279–285. [20] Chang, P.-T.; Chang, C.-H. An integrated artificial intelligent computeraided process planning system. International Journal of Computer Integrated Manufacturing 2000. 13(6): 483–497. [21] Amaitik, S.; Kiliç, S. An intelligent process planning system for prismatic parts using STEP features. The International Journal of Advanced Manufacturing Technology 2007. 31(9-10): 978–993. [22] Basheer, I.; Hajmeer, M. Artificial neural networks. Fundamentals, computing, design, and application. Journal of Microbiological Methods 2000. 43(1): 3–31. [23] Kubat, M.: An Introduction to Machine Learning, Cham: Springer International Publishing, 2015. [24] Joo, J.; Park, S.; Cho, H. Adaptive and dynamic process planning using neural networks. International Journal of Production Research 2001. 39(13): 2923–2946. [25] Reinhart, G.; Geiger, F. Adaptive Maschinenbelegungsplanung auf Basis produktspezifischer Entstehungsdaten. Zeitschrift für Wirtschaftlichen Fabrikbetrieb 2011. 106(6): 428–432. [26] Monostori, L.; Kádár, B.; Bauernhansl, T.; Kondoh, S.; Kumara, S.; Reinhart, G.; Sauer, O.; Schuh, G.; Sihn, W.; Ueda, K. Cyber-physical systems in manufacturing. CIRP Annals - Manufacturing Technology 2016. 65(2): 621–641. [27] Denkena, B.; Schmidt, J.; Krüger, M. Data Mining Approach for Knowledge-based Process Planning. Procedia Technology 2014. 15: 406– 415. [28] Fayyad, U.; Piatetsky-Shapiro, G.; Smyth, P. From Data Mining to Knowledge Discovery in Databases. AI magazine 1996. 17(3): 37–54. [29] Schuh, G.; Potente, T.; Thomas, C.; Brambring, F. Approach for Reducing Data Inconsistencies in Production Control. In: Zaeh, M. F., ed. Enabling Manufacturing Competitiveness and Economic Sustainability, Cham: Springer International Publishing, 2014; p. 347–351. [30] Codd, E.; Codd, S.; Salley, C. Providing OLAP (on-line analytical processing) to user-analysts. An IT mandate. Codd and Date 1993. 32. [31] Inmon, W. H. Building the data warehouse, 3rd ed. New York: J. Wiley, 2002. [32] Witten, I. H.; Frank, E.; Hall, M. A. Data mining: practical machine learning tools and techniques. 3rd ed. Burlington: Elsevier; 2011