Design of a New Automated Fault Detector based on artificial intelligence and Big Data Techniques

Design of a New Automated Fault Detector based on artificial intelligence and Big Data Techniques

ScienceDirect ProcediaScienceDirect Computer Science 00 (2019) 000–000 Available online at www.sciencedirect.com Available online at www.sciencedire...

1MB Sizes 0 Downloads 59 Views

ScienceDirect ProcediaScienceDirect Computer Science 00 (2019) 000–000

Available online at www.sciencedirect.com

Available online at www.sciencedirect.com

www.elsevier.com/locate/procedia

ScienceDirect

www.elsevier.com/locate/procedia

Procedia Computer Science 00 (2019) 000–000

Procedia Computer Science 163 (2019) 460–471

16th International Learning & Technology Conference 2019 16th International Learning & Technology Conference 2019

Design of a New Automated Fault Detector based on artificial intelligence and Big Data Techniques Design of a New Automated Fault Detector based on artificial andM. Big Data Kamel intelligence H. Rahouma, Farag Afify and Techniques Hesham F. A. Hamed Electrical Engineering Department, Faculty of Engineering, Minia University, Minia, Egypt.

Correspondence email: Kamel H. Rahouma, Farag [email protected] Afify and Hesham F. A. Hamed Electrical Engineering Department, Faculty of Engineering, Minia University, Minia, Egypt. Correspondence email: [email protected]

Abstract Concerns over energy and environmental issues around the world have led to the worldwide focus on energy use reduction. Abstract Studies in each area of energy have shown that residential and commercial construction sectors consume more power than other sectors such industry, services, transportation. Studies energy consumption in building sectors have Concerns overasenergy andagriculture, environmental issues and around the world have led of to the worldwide focus on energy use reduction. reportedinthat energy of have 10% shown to 30%that canresidential be obtained bycommercial using artificial intelligence (AI),consume the system would be capable of Studies each area savings of energy and construction sectors more power than other detectingsuch and as analyzing anomalies in energy usageand pattern assessing, diagnosing suggesting the bestinsolution in suitable sectors industry, agriculture, services, transportation. Studies of and energy consumption building sectors time. have This paper proposes integrate and hybridize between AI techniques big data algorithms which enhance reported that energy to savings of 10% to 30% can be obtained by using and artificial intelligence (AI), the can system wouldmonitoring be capableand of controllingand building systems, increasing comfort decreasing efficiently the running costs. Intheaddition, the authors suggest detecting analyzing anomalies in energy usageand pattern assessing, diagnosing and suggesting best solution in suitable time.a tool which aims to automatically abnormal energy consumption AI and bigwhich data can which are produced by and the This paper proposes to integrate anddetect hybridize between AI techniques andby bigusing data algorithms enhance monitoring Building Management Systemincreasing (BMS). This happens by designing a software application that called Fault Detection Toola controlling building systems, comfort and decreasing efficiently the running costs. In is addition, the authors suggest (FDT) which automatically detects the abnormalities of energy consumption, thedata use which of different resources tool which aims to automatically detect abnormal energy consumption by usingoptimizes AI and big are produced by and the analyzes faults, complaints and (BMS). time taken terminatebythem. Experimental results show that with the proposed approach, Tool it is Building Management System Thisto happens designing a software application that is called Fault Detection possiblewhich to accurately detectdetects anomalous patterns in building energy consumption. This tool will ofbedifferent a part ofresources an artificial (FDT) automatically the abnormalities of energy consumption, optimizes the use and intelligentfaults, decision-making analyzes complaints system. and time taken to terminate them. Experimental results show that with the proposed approach, it is possible to accurately detect anomalous patterns in building energy consumption. This tool will be a part of an artificial intelligent decision-making system. © 2019 The Authors. Published by Elsevier B.V. © 2019 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0/) This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0/) © 2019 The Authors. Published by Elsevier B.V. committee of the 16th International Learning & Technology Conference 2019. Peer-review responsibility ofthe thescientific scientific Peer-review under under responsibility of committee of the 16th International Learning & Technology Conference 2019. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0/) Peer-review of theabnormalities; scientific committee ofFault the 16th International Learning & Technology Conference 2019. Keywords: AI;under BMS;responsibility Energy consumption Automated Detection Keywords: AI; BMS; Energy consumption abnormalities; Automated Fault Detection

1. Introduction

1. Introduction During the day when you are reading this, more data will be produced than the volume of Data contained in all printed material in libraries. The revolution in the generation of massive data amounts comes along with Internet During theallows day when you are reading this, moreelectronic data will devices be produced than the[1] volume of Data contained in all usage which data exchange between various and humans. printed material in libraries. The revolution in the generation of massive data amounts comes along with Internet usage which allows data exchange between various electronic devices and humans. [1] 1877-0509 © 2019 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0/) Peer-review the scientific committee 1877-0509 ©under 2019responsibility The Authors. of Published by Elsevier B.V.of the 16th International Learning & Technology Conference 2019. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0/) Peer-review under responsibility of the scientific committee of the 16th International Learning & Technology Conference 2019.

1877-0509 © 2019 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0/) Peer-review under responsibility of the scientific committee of the 16th International Learning & Technology Conference 2019. 10.1016/j.procs.2019.12.129

10

Kamel H. Rahouma et al. / Procedia Computer Science 163 (2019) 460–471 Kamel Rahouma and Farag Afify / Procedia Computer Science 00 (2019) 000–000

461

Big data is a term that represents the large quantities of information both structured plus unstructured resulting from a lot of operations in all fields. By using big data to penetrations that guide to good decisions and improving a lot of operations. [2] Big data concept when data volumes, number of operations plus the number of data sources are so huge and difficult that they require special processes and Techniques in order to analyzed, stored and collect analyzed.[3] This also forms the base for the commonly used description of big data, this three V: Variety, Velocity, and Volume as presented in Figure 1.

Fig. 1. The three V of Big Data.

- Volume; Huge quantities of information, from datasets with sizes of terabytes to zettabyte. - Velocity; Huge quantities of data from transactions with large refresh rate resulting in data streams coming at high speed and the time to act on the basis of those information streams will usually be very short. There appears a shift of batch processing to real-time streaming. [4, 5] - Variety; Data come from different data sources. For the beginning, data can come off both external and internal information source. More importantly, data can come in various formats like transaction and log data from various applications, semi-structured data like XML data, structured data as a database table, unstructured data like audio, video, text, images, streams, and more. It notes that the production of unstructured data has been increasing recently. [6, 7, 8] From SIRI [9] to AlphaGo [10], AI is improving quickly. While most people usually think of AI as robots with human-like features, AI can include anything like smart sensors, industry sectors, e-Commerce prediction algorithms, diagnosis, and detection of faults, call centers and etc. These systems have a direct effect on the overall economy in the world and achieve considerable on reducing cost. Most AI applications today are designed to do specific jobs (e.g., facial recognition, internet searches, monitoring, and control, doing certain tasks in the industry). While this application may exceed people at a specific job, for example solving equations or playing chess, general AI would exceed people at almost every cognitive task. In recent years, the United States of America has supported primary research on AI, which is focused on automation and pattern identification (sound, pictures, etc.). Microsoft has published real-time translation robots and innovative image identification technologies [11]. Amazon employs AI for independent robots in delivery operations [12]. Facebook has also improved facial identification techniques based on AI named “DeepFace” [13]. Stanford University developed the robot car by using AI and achieved faster time than an active racer [14]. “Big Data” and “artificial intelligence” have captured interest Researchers, Manufacturers, merchants, and all people. The systems which use these techniques play a vital role in the economic, political and social sphere. This paper consists of 6 sections. Section (1) is an introduction and section (2) is a background. Section (3) is a literature review and section (4) illustrates the proposed system. Section (5) introduces the obtained results of the system and section (6) highlights some conclusions. At the end of the paper, there is a list of the used references.

462

Kamel H. Rahouma et al. / Procedia Computer Science 163 (2019) 460–471 Kamel Rahouma and Farag Afify / Procedia Computer Science 00 (2019) 000–000

3

2. Background In this section, we introduce some of the basic issues in regard to energy consumption and how to use AI techniques to optimize it. 2.1. Integration between Big data and AI: AI makes it possible for machines to learn from experience, arrange to new inputs and do human-like jobs. Most AI applications like chess-playing computers, self-driving cars Passed through many stages of deep learning and natural language processing. Using those technologies, machines can be trained to perform special tasks by processing large amounts of information and identifying patterns in this information. [15] Big data [16] reflects the practices of combining huge volumes of diversely sourced information often using artificial intelligence applications, to provide insight. One of the biggest benefits of big data is derived from the monitoring of human behaviour, collectively and individually, and its predictive potential. [17] The relationship between artificial intelligence and big data is bi-directional: AI needs a huge amount of information to learn. In the other direction, big data applies AI techniques in extracting value from big datasets. One of the main issues relating to large data is information to individuals: transparency. Unless people are given with suitable data and control, they 'will be subject to arrangements that they do not understand and have no control over. [18] There is a very important concern with regards to artificial intelligence and big data is the bias expressed through the input dataset provided for training the artificial intelligence. As the machine learns from the information provided and has no means to contrast that information with a larger picture, whatever bias is contained in the training set will influence the predictions made. [19, 20] 2.2. Artificial Neural Network (ANN) Concept: ANN is a machine learning technique that has been popularly used in health, finance, engineering, and science for predicting the effects of input variables on outputs, by using a weighing system that adjusts the networks, to decrease the errors to the lowest possible rate. [21]There are three main layers in ANN which called the input layer, hidden layer, and output layer, which are linked with each other. They also have weighted input ingredients that are changed as the signals pass through the hidden neurons, which give their outputs using the sigmoidal function Equation (1) [22]. The output weight produced by the hidden neurons (hi), which are linked to the input neurons in adjacent layers and connected to the output neurons with a weight factor, can be estimated with Equation (2) [23,24]



ሼ

 





(1)  ∑    

Where XiS represent the output of node i in layer s, Xj s-1 represents the output of node j in layer s – 1 and si represents the weighted sum (wij) of the inputs to node i.  (∑ 

ሻ (2)

Here, ơ(.) is the activation function; N is the number of input neurons; ʚij is the weights between input neuron j and hidden neuron i; xj is the input values to the input neurons; and Ti hid represents the threshold term of the hidden neuron. In ANN network training, the weights are adjusted continually, to reduce the difference (") between the desired value and the target value to the bare minimal, per Equation (3) [24].  ∑∑ 

ሺሻ

(3)

Here, mt, mo, Yij and Dij represent the number of training samples, the number of output nodes of the training samples, the output of the training network and the desired value of the target components (response), respectively.

4

Kamel H. Rahouma et al. / Procedia Computer Science 163 (2019) 460–471 Kamel Rahouma and Farag Afify / Procedia Computer Science 00 (2019) 000–000

463

Energy issues become a global issue and there is a growing interest BMS which efficiently manages and controls the energy used in commercial, residential buildings, and hospitalities. BMS is an optimal control system that can rationally manage energy use in the building to create a pleasant indoor environment while minimizing energy consumption. BMS achieves energy efficiency by collecting and monitoring data from operation sensors (for example lighting, temperature, and humidity) and life pattern data of residents. [25, 26] BMS is a necessary part in the operation of any modern business building, especially the buildings which name smart buildings. Owners and operators generally maintain and or upgrade these controls systems with the following aims: To ensure the essential level of operation is maintained and that the intended design conditions are met, to improve efficiency, reduce their energy consumption and CO2 emissions, to minimize risk and complaints handling and control system useful life. Technology has delivered extensive improvements in cost, capability, and reliability to both the hardware and software platforms associated with building control systems. As illustrated in Figure 2, traditional BMS is software running on a server located in the building. Standard BMS contains three levels of functions: (1) The field layer contains sensors, actuators, and devices connecting to the physical world; (2) the automation layer applies strategies derived from a set of rules, basics, and parameters; (3) the management layer configures and manages the other layers. [27, 28] Old BMS usually offer limited capacities in terms of sensor data storage and act as isolated systems often unreachable from the outside but new BMS developed to contain remote solutions for storage and data processing. In addition to the ability to scale up at the level of buildings operations, this development brings new opportunities in terms of BMS development through information storage, Massive development in sensors, and applications levels. This opening is supported by the development of the open and standardized principles of the Internet of Things (IoT) and Web of Things. [29] It allows for seamless integration of new data sources, like weather prediction available in web services or mobile sensors. It also allows using a modern web application. Finally, the availability of long term historical data allows using advanced mathematical models and machining learning technique. [30]

Fig. 2. BMS Structure and big data platforms.

In developed countries, residential and business buildings are consuming more power than other sectors, for example, service, transportation, and industry sectors. For example, in the United States, power consumption by buildings accounts for more than 40% of total national energy consumption. Furthermore, it is predicted that the total power consumption in the building sectors will be increased in the future, in all over world. [31] Everywhere buildings now produce a huge amount of information on power consumption, operations, and activities through systems such as BMS and meters (like sub-meters and smart meters). However, the majorities of data are not utilized and are thrown away. by using big data and AI techniques can play an important role in utilizing this information and accurately evaluating how power in buildings is consumed and what can be done to save power, improving energy efficiency, and decrease GHG emissions. Data size in BMS has increased dramatically, which results in gaps and challenges, in BMS are difficult to be

464

Kamel H. Rahouma et al. / Procedia Computer Science 163 (2019) 460–471 Kamel Rahouma and Farag Afify / Procedia Computer Science 00 (2019) 000–000

5

handled within a tolerable operating time and hardware resources. Therefore, the big data technology has been introduced to the BMSs like other fields of life. Engineering systems ranging of all kinds and sizes are exposed to various kinds of faults. Faults may produce sub-optimal operation and decline in performance if not even preventing the whole system from running. It is, therefore, necessary to detect faults immediately and to know their conditions, hardness, and results. FDD tools are proposed to address these problems. The system is defined as faulty if one of the characteristic features or parameters of the system changes from a normal state. It is clear that in a normal system all parameters and characteristic features are in an agreeable ordinary and standard form. [32] FD is a subfield of control engineering which contains a lot of techniques that help in monitoring a system, identifying when a fault has happened and pinpointing the kind of fault and its place. Two strategies can be identified: direct pattern identification of sensor readings that show a fault and an analysis of the difference between the sensor readings and expected values, obtained from some model. In another case, it is typical that a fault is declared to be identified if the discrepancy or residual goes over a settled threshold. It is then the task of fault isolation to classify the kind of fault and its place in the system. [33-36]. While most of the power goes into primary building use, a significant quantity of energy is lost due to malfunctioning building system equipment and wrongly configured BMS. For example, wrongly configured set points or building equipment, or misplaced sensors and actuators, can contribute to deviations of the real energy consumption from the predicted one. Our proposed is motivated by these posed challenges and aims at pinpointing the kinds of errors in the BMS elements that can have an impact on the power efficiency of a building, as well as examining the processes that can be utilized for their identification and diagnosis. In this paper, we propose the technique which aims to automatically detect abnormal energy consumption by leveraging BMS big data and AI techniques follow up complaints and time used to finish them. 3. Previous work: In [37] a new methodology for the determination of different behaviour patterns was presented, where each behaviour pattern represents a set of similar daily profiles and occurs in the building (building system) with different frequencies. The methodology is based on two clustering procedures, fulfilled in parallel, and has been implemented using MATLAB software with the help of artificial neural networks. In [38] Platform was proposed, a distributed system for storing and processing building data. The platform enables new potentials in terms of data analytics and applications development, with the ability to scale up seamlessly from one smart building to several, in the direction of smart areas and smart cities. In [39] A novel system was presented capable of automating detection and diagnosis of faults in commercial building HVAC systems. This system is able to detect faults accurately and robustly and in real-time using data from an operational building in Newcastle, Australia, and in a standard ASHRAE 1020 project’s FDD dataset. An FDD technique was proposed using Hidden Markov Models has been produced to learn probabilistic relationships between collections of points during both normal and faulty operation. This can passively infer the likelihood of similar patterns in the information during future operation with a great degree of efficiency. 4. Proposed work: We have presented a new tool capable of automating detection and diagnosis of faults by using integration between AI and BMS big data. This tool is able to detect faults accurately and robustly and in real-time using new and historical BMS data. FDT (Fault Detection tool) is a software application doing a specific task for detecting abnormal energy consumption by using BMS big data analytics. BMS collects huge quantities of data, such as operational data (eg, temperature, electricity, air conditioning, and lighting), energy usage pattern data, and weather data. These datasets have features of time series. Therefore, FDT uses the temporal database model to effectively model the data set collected by BMS. FDT is built on the top of BMS. As shown in figure 3

6

Kamel Rahouma and Farag Afify / Procedia Computer Science 00 (2019) 000–000 Kamel H. Rahouma et al. / Procedia Computer Science 163 (2019) 460–471

465

Fig. 3. An overview of the proposed FDT for detecting abnormal energy consumption.

4.1. Fault Detection with ANN: To apply fault detection the current system status, historic and new big data of the inputs are needed because the behaviour of the system will be different about the historic pattern. The procedure used for this process is shown in Figure 4.

Fig. 4. The integrated process of fault detection with AI.

466

Kamel H. Rahouma et al. / Procedia Computer Science 163 (2019) 460–471 Kamel Rahouma and Farag Afify / Procedia Computer Science 00 (2019) 000–000

7

The first step of real-time FDT is pre-processing, which needs the replacement of missing values, delete of incomplete columns and rows and extreme values. These steps of information cleaning can also include data integration, reduction, discretization, and transformation to make the tool work fast and forbid bogus results. Thus, redundant input variables such as those that were constant were deleted, and missing values were replaced with zeros and by averaging the nearest neighbours’ values of the missing value cells. This research used importance scoring to establish the influences of the source elements (input data) on the target element’s (output data) behaviour. These steps were meant to determine actionable data processing size that will have an effective contribution to the attitude of the target component. Hence, the first process in the predictive analytics was to correlate the readings of the source elements with that of the target element and the cumulative influence of the source elements on the target element recorded. The ANN model training, which estimated the value of the target component using the combination of the information in Equations (1)–(3), was executed using the structure shown in Figure 5 in a cross-validation ensemble procedure.

Fig. 5. ANN architecture. ANN structure used for FD improvement, Tx1 is the target element (output) and Tx2, Tx3, …, Tx234 represent the source elements (inputs) of the system. - The suggested method uses two sets of input data, the current BMS data, and the past BMS data, and calculates similarities between them to determine if there is an abnormality in the current building energy usage pattern. Hence, the suggested FDT consists of two subtasks: ((1) fetch historical data from the datasets, (2) determine the similarity between the current BMS reading values and past values. Building systems perform very differently in different weather and operating modes. In order to efficiently detect abnormal patterns in building power usage, it is important to compare the observed power consumption patterns under similar environmental and operational conditions. Thus, the first sub-task of the suggested FDT is to detect power usage information observed in past circumstances similar to the current circumstances.

8

Kamel Rahouma and Farag Afify / Procedia Computer Science 00 (2019) 000–000 Kamel H. Rahouma et al. / Procedia Computer Science 163 (2019) 460–471

4.2. The algorithm: 1- Fitch inside data now. 2- Fitch inside temp data a year ago. 3- Validate data and remove extreme and missed columns throw ANN. 4- Compare each parameter value between two sets of data. 5- If all deference is less than or equal to 5% show OK message. And close. 6- Fitch outside values of the parameter (K) now. 7- Fitch outside values of the parameter (K) a year ago. 8- Compare both values. 9- If deference is greater than 5% show OK message. 10- If deference is less or equal to 5% show error and saves the report. 4.3. Algorithm application: // get bms data from big data D1> Key , value< = Fitch new BMS data (inside) D2> Key , value< = Fitch old BMS data (inside) D3> Key , value< = Fitch new BMS data (outside) D4> Key , value< = Fitch old BMS data (outside) // Validate the data values throw ANN. D1> Key , value< = validate_data (D1) D2> Key , value< = validate_data (D2) D3> Key , value< validate_data (D3) D4> Key , value< = validate_data (D4) // loop for each value in new collection For each K in get keys (D1) do // Get old and new values (Inside) for current parameter. V1= get value (D1,K); V2= get value (D2,k); // calculate Deff % for inside values Deff1= V1-V2 ÷ V1*100; // If deff > 5% check outside values If deff1 > 5% then // Get old and values (outside) for current parameter V3= get value (D3,K); V4= get value (D4,k); // calculate Deff% for outside Deff2= V3-V4 ÷ V3*100; // if deff > 5 % trigger alarm If Deff2 <= 5% then Alarm (K) and store result; End if End if End For

4.4. Complexity analysis of the algorithm: By assuming that each set contain N pairs of data, we find: // get bms data from big data D1> Key , value< = Fitch new BMS data (inside) >>> take N time units, N Memory Unit D2> Key , value< = Fitch old BMS data (inside) >>> take N time units, N Memory Unit D3> Key , value< = Fitch new BMS data (outside) >>> take N time units, N Memory Unit D4> Key , value< = Fitch old BMS data (outside) >>> take N time units, N Memory Unit >>> Take 4N time Unit and 4N Memory Unit // Validate the data values throw ANN. D1> Key , value< = validate_data (D1) >>> take N time units D2> Key , value< = validate_data (D2) >>> take N time units

467

468

Kamel H. Rahouma et al. / Procedia Computer Science 163 (2019) 460–471 Kamel Rahouma and Farag Afify / Procedia Computer Science 00 (2019) 000–000

9

D3> Key , value< validate_data (D3) >>> take N time units D4> Key , value< = validate_data (D4) >>> take N time units >>> Take 8N time Unit and 4N Memory Unit // loop for each value in new collection This loop will repeated N times. For each K in get keys (D1) do >>> take 1 time unit, take k memory unit // Get old and new values (Inside) for current parameter. V1= get value (D1, K); >>> take 1 time unit, take v memory Unit V2= get value (D2, k); >>> take 1 time unit, take v memory Unit // calculate Deff % for inside values Deff= |V1-V2| ÷ V1*100; >>> take 1 time unit, take X memory Unit (0 -100%) // If deff > 5 % check outside values If deff > 5% then >>> take 1 time unit, take X memory Unit for (0 -100%) // Get old and values (outside) for current parameter V3= get value (D3,K); >>> take 1 time unit, take v memory Unit V4= get value (D4,k); >>> take 1 time unit, take v memory Unit // calculate Deff% for outside Deff= |V3-V4| ÷ V3*100; >>> take 1 time unit // if deff > 5 % trigger alarm If Deff <= 5% then >>> take 1 time unit, Alarm (K) and store result; >>> take 1 time unit End if >>> take 1 time unit End if >>> take 1 time unit End For >>> take 1 time unit This program will take 8N + 13N = 21N time units This program will take 4N + k + 4V + 2X memory units

5. Experimental results: In this section, we describe the experiments carried out to evaluate the effectiveness of the suggested approach. We execute the algorithm twice. Each time with a different dataset. This is shown in the following subsections. 5.1. First dataset test: The first data sets we test tool by it, are generated by EnergyPlus [40] which is the most popular energy simulation tool. For evaluation purpose, we create two different data sets: the past power usage data set and the current energy usage data set. To collect the past energy usage data set, a residential building was simulated for one year on EnergyPlus. Past power usage data was collected is modelled and stored in XML files. For the evaluation of the suggested method, we also generated the current power usage data set (which compatible with the testing data set). By simulating the same building on EnergyPlus.

Fig. 6. FDT result: normal and anomalous energy consumption.

Furthermore, some power usage data were randomly extracted and added noise. Note that the idea we added noise to the power usage data produced by the EnergyPlus is to create a set of test data that works as anomalous power usage patterns. Figure 6 shows, detecting anomalous patterns in building power usage and reporting alarms to the



1 0operation

Kamel H. Rahouma et al. / Procedia Computer Science 163 (2019) 460–471

469

Kamel Rahouma and Farag Afify / Procedia Computer Science 00 (2019) 000–000

centre in real time to solve these problems. As can be seen with FDT, we can achieve very low error rate. This verifies that the proposed approach is very effective in detecting anomalous patterns in building energy consumption.

5.2. The second dataset test: The experimental use case for illustrating the effective of FDT is selected on the basis of BMS used for monitoring and controlling the operation of a great building like an airport. The experimental data is composed of inputs from 7 halls, 154 offices, 12 data room, and 22 power room which belong to one of the buildings which selected for this task. Each place is equipped with local air-conditioning (AC) units, which can be managed by a control panel located inside the room. Users are enabled to switch AC unit on or off and change desired room temperature. AC unit controls the speed of the fan which supplies cold air into the room, opens or closes valves of central heating radiators. AC unit automatically switches off if windows in the place are opened. All AC operations have a direct effect on energy consumption. The total size of the dataset is 30480 XML file produced from BMS contains all data of selected places. After using FDT we found a great impact on energy consumption and a lot of energy use distortions have been reported and treated.

Fig. 7. Fault detection phase. After applying FDT any distortion in power consumption, the alarm will be sent to the operation centre to take all procedure for return to the normal state as shown in figure 7. 6. Conclusions: With the growing concern about energy, the need for improving its efficiency has become a very important topic all over the world. It is important to save energy consumption and to detect abnormal consumptions. In this paper, we first review and study the integration of big data and AI. Furthermore, a new tool was proposed to detect anomalous power consumptions by exploiting BMS big data and AI techniques, when abnormal energy usage is detected. This can enhance monitoring and controlling building systems, increasing comfort and decreasing efficiently the

470

Kamel H. Rahouma et al. / Procedia Computer Science 163 (2019) 460–471 Kamel Rahouma and Farag Afify / Procedia Computer Science 00 (2019) 000–000

11

running costs. In addition, the authors proposed a tool which aims to automatically detect abnormal energy consumption by using AI and big data which are produced by the Building Management System (BMS). An automated Fault Detection Tool (FDT) was designed and applied to detect the abnormalities of energy consumption, optimizes the use of different resources and analyzes faults, complaints and time have taken to terminate them. Experimental results showed that with the proposed approach, it is possible to accurately detect anomalous patterns in building energy consumption. This tool is a part of an artificial intelligent decision-making system. References [1] John Gantz and David Reinsel. The digital universe in 2020: Big Data, Bigger Digital Shadows, and Biggest Growth in the Far East. Tech.

rep. Internet Data Center (IDC), 2012. http://www.emc.com/collateral/analyst-reports/idc-the-digital-universe-in-2020.pdf.

[2] Marcos D. , Rodrigo Calheiros , Silvia Bianchi , Marco A.S. Netto and Rajkumar Buyyab " Big Data computing and clouds: Trends and

future directions" INRIA, LIP, ENS de Lyon, France, The University of Melbourne, Australia and IBM Research, Brazi, J. Parallel Distrib. Comput. 79–80, Elsevier 2015. [3] Florin Pop and Joanna Kołodziej “Resource Management for Big Data Platforms and Applications” Institute of Computer Science, Cracow University of Technology, Poland, a book in “Studies in Big Data” Springer series http://www.springer.com/series/11970- 2016. [4] Mark A. Beyer and Douglas Laney. “The Importance of 'Big Data': A Definition”. Gartner, 2012. [5] Bill Franks. “Taming the big data tidal wave”. Wiley, 2012. [6] David R. Hardoon and Galit Shmueli. “Getting started with business analytics – insightful decision making”. Talor & Francis Group.2013. [7]Foster Provost and Tom Fawcett. “Data science for business”. O’Relly, 2013 [8] Thomas H. Davenport and D.J. Patil . “Data Scientist: The Sexiest Job of the 21st Century”, Harvard Business Review, 2012. [9] Siri, https://en.wikipedia.org/wiki/Siri (Accessed on 2018/12). [10] AlphaGo, https://deepmind.com/research/alphago/ (Accessed on 2018/12). [11] Microsoft Translator Speech API, https://www.microsoft.com/en-us/translator/speech.aspx (Accessed on 2018/12). [12] Amazon Prime Air, https://www.amazon.com/Amazon-Prime-Air/b?node=8037720011 (Accessed on 2018/12). [13] Y. Taigman, M. Yang, M. Ranzato, L. Wolf, “DeepFace: Closing the Gap to Human-Level Performance in Face Verification,” IEEE International Conference on Computer Vision and Pattern Recognition (CVPR2014), pp.1-8, 2014. [14] Stanford Artificial Intelligence Laboratory, http://ai.stanford.edu/ (Accessed on 2018/12). [15 ] https://www.sas.com/en_us/insights/analytics/what-is-artificial-intelligence.html (Accessed on 2018/12). [16] EDPS Meeting the challenges of big data, A call for transparency, user control, data protection by design and accountability (19 November 2015) https://secure.edps.europa.eu/EDPSWEB/webdav/site/mySite/shared/Documents/Consultation/Opinions/2015/15-11-19_Big_Data_EN.pdf [17] EDPS Towards a new digital ethics: Data, Dignity and Technology (11 September 2015) https://secure.edps.europa.eu/EDPSWEB/webdav/site/mySite/shared/Documents/Consultation/Opinions/2015/15-09-11_Data_Ethics_EN.pdf [18] WP29 Statement of the WP29 on the impact of the development of big data on the protection of individuals with regard to the processing of their personal data in the EU (16 September 2014) - http://ec.europa.eu/justice/data-protection/article-29/documentation/opinionrecommendation/files/2014/wp221_en.pdf [19] Barbini, L.; Ompusunggu, A.P.; Hillis, A.J.; du Bois, J.L.; Bartic, A. Phase editing as a signal pre-processing step for automated bearing fault detection. Mech. Syst. Signal Process. 2017, 91, 407–421. [20] Kumar, A.; Shankar, R.; Thakur, L.S. A big data driven sustainable manufacturing framework for condition-based maintenance prediction. J. Comput. Sci. 2017. [21] Manco, G.; Ritacco, E.; Rullo, P.; Gallucci, L.; Astill, W.; Kimber, D.; Antonelli, M. Fault detection and explanation through big data analysis on sensor streams. Expert Syst. Appli. 2017, 87, 141–156. [22] Cai, J.; Cottis, R.A.; Lyon, S.B. Phenomenological modelling of atmospheric corrosion using an artificial neural network. Corros. Sci. 1999, 41, 2001–2030. [23] Wang, S.C. Artificial neural network. In Interdisciplinary Computing in Java Programming; Springer: New York, NY, USA, 2003; 81–100. [24] Wang, P.; Vachtsevanos, G. Fault prognostics using dynamic wavelet neural networks. AI EDAM 2001, 15, 349–365. [25] Residential and commercial buildings. U.S. Energy. Information Administration, International Energy Outlook 2016. [26] Seerig, A., Zakovorotnyi, A., Combination of Monte-Carlo Approach and Artificial Neural Network for the Determination of Typical Swiss Building Load Profiles. Conference proceedings of the Building Simulation Conference “BauSIM 2016”, Dresden, Germany: 377-381. [27] Bovet, G.. A scalable and sustainable web of buildings architecture. Ph.D. thesis; Telecom ParisTech; 2015. [28] Bovet, G., Hennebert, J.. A web-of-things gateway for knx networks. In: Smart SysTech 2013; European Conference on Smart Objects, Systems and Technologies. 2013, p. 1–8. [29] Bovet, G., Hennebert, J.. Offering web-of-things connectivity to building networks. In: Proceedings of the 2013 ACM Conference on Pervasive and Ubiquitous Computing Adjunct Publication. UbiComp ’13 Adjunct; New York, NY, USA: ACM; 2013, p. 1555–1564. [30] Ridi, A., Zarkadis, N., Bovet, G., Morel, N., Hennebert, J. Towards reliable stochastic data-driven models applied to the energy saving in buildings. In: International Conference on Cleantech for Smart Cities & Buildings from Nano to Urban Scale (CISBAT 2013). 2013, p. 501–506. [31] American Power Conversion (APC) - https://www.apc.com (Accessed on 2018/12). [32] Mardiana A and Riffat SB `` Building Energy Consumption and Carbon dioxide Emissions: Threat to Climate Change``Journal of oJ Earth Science & Climatic Change, January 2015. [33] Z. Gao, C. Cecati, and S. X. Ding, ``A survey of fault diagnosis and fault-tolerant techniques Part I: Fault diagnosis with model-based and signal-based approaches,'' IEEE Trans. Ind. Electron., vol. 62, no. 6, pp. 3757_3767, Jun. 2015. [34] K. Tidriri, N. Chatti, S. Verron, and T. Tiplica, ``Bridging data-driven and model-based approaches for process fault diagnosis and health monitoring: A review of researches and future challenges,'' Annu. Rev. Control, vol. 42, pp. 63_81, Sep. 2016. [35] Y.-K. Liu, G.-H. Wu, C.-L. Xie, Z.-Y. Duan, M.-J. Peng, and M.-K. Li, ``A fault diagnosis method based on signed directed graph and matrix for nuclear power plants,'' Nucl. Eng. Design, vol. 297, pp. 166_174, Sep. 2016.

Kamel H. Rahouma et al. / Procedia Computer Science 163 (2019) 460–471 471 1 Kamel Rahouma and Farag Afify / Procedia Computer Science 00 (2019) 000–000 2 [36] B. Cai, Y. Zhao, H. Liu, and M. Xie, ``A data-driven fault diagnosis methodology in three-phase inverters for PMSM drive systems,'' IEEE Trans. Power Electron., vol. 32, no. 7, pp. 5590_5600, Jul. 2017. [37] Andrii Zakovorotnyi, Axel Seerig " Building energy data analysis by clustering measured daily profiles" CISBAT 2017 International Conference – Future Buildings & Districts – Energy Efficiency from Nano to Urban Scale, CISBAT 2017 6-8 September 2017, Lausanne, Switzerland. [38] Lucy Linder, Damien Vionnet, Jean-Philippe and Jean Hennebert " Big Building Data- a Big Data Platform for Smart Buildings " CISBAT 2017 International Conference – Future Buildings & Districts – Energy Efficiency from Nano to Urban Scale, CISBAT 2017 6-8 September 2017, Lausanne, Switzerland. [39] Adam Kučera,Petr Glos and Tomáš Pitner " Fault Detection in Building management system networks " 12th IFAC Conference on Programmable Devices and Embedded Systems, the International Federation of Automatic Control September 25-27, 2013. Velke Karlovice, Czech Republic. [40] EnergyPlus, https://energyplus.net/ (Accessed on 2018/12).