Ứng dụng deep learning để nhận dạng lỗi trong hệ thống công nghiệp

BỘ GIÁO DỤC VÀ ĐÀO TẠO TRƯỜNG ĐẠI HỌC SƯ PHẠM KỸ THUẬT THÀNH PHỐ HỒ CHÍ MINH LUẬN VĂN THẠC SĨ PHẠM HUỲNH THẾ ỨNG DỤNG DEEP LEARNING ĐỂ NHẬN DẠNG LỖI TRONG HỆ THỐNG CÔNG NGHIỆP NGÀNH: KỸ THUẬT ĐIỆN Tp Hồ Chí Minh, tháng 11/2022 BỘ GIÁO DỤC VÀ ĐÀO TẠO TRƯỜNG ĐẠI HỌC SƯ PHẠM KỸ THUẬT THÀNH PHỐ HỒ CHÍ MINH LUẬN VĂN THẠC SĨ PHẠM HUỲNH THẾ ỨNG DỤNG DEEP LEARNING ĐỂ NHẬN DẠNG LỖI TRONG HỆ THỐNG CÔNG NGHIỆP NGÀNH: KỸ THUẬT ĐIỆN TỬ - 8520203 Hướng dẫn khoa học: PGS.TS TRƯƠNG ĐÌNH NHƠN Tp Hồ Chí Minh, tháng 10/2022 LUẬN VĂN TỐT NGHIỆP i LUẬN VĂN TỐT NGHIỆP LUẬN VĂN TỐT NGHIỆP LUẬN VĂN TỐT NGHIỆP LUẬN VĂN TỐT NGHIỆP LUẬN VĂN TỐT NGHIỆP LUẬN VĂN TỐT NGHIỆP i LUẬN VĂN TỐT NGHIỆP LÝ LỊCH KHOA HỌC I LÝ LỊCH SƠ LƯỢC: Họ & tên: Phạm Huỳnh Thế Giới tính: Nam Ngày, tháng, năm sinh: 22/06/1995 Nơi sinh: Bình Thuận Quê quán: Mộ Đức, Quảng Ngãi Dân tộc: Kinh Chỗ riêng địa liên lạc: Căn hộ A10-10 Saigon Gateway, 702 Xa lộ Hà Nội, P Hiệp Phú, TP Thủ Đức, TP HCM Điện thoại quan: Điện thoại nhà riêng: Fax: E-mail:phamhuynhthe@gmail.com II QUÁ TRÌNH ĐÀO TẠO: Đại học: Hệ đào tạo: Chính quy Thời gian đào tạo từ 09/2013 đến 09/2017 Nơi học (trường, thành phố): Trường ĐH Sư Phạm Kỹ Thuật TP HCM, TP.Hồ Chí Minh Ngành học: Cơng Nghệ Kỹ Thuật Điện tử, Truyền thông Tên đồ án, luận án môn thi tốt nghiệp: Sử dụng IoT giám sát sức khỏe Ngày & nơi bảo vệ đồ án, luận án thi tốt nghiệp: 07/2018 Trường ĐH Sư Phạm Kỹ Thuật TP HCM i LUẬN VĂN TỐT NGHIỆP 68 LUẬN VĂN TỐT NGHIỆP 69 LUẬN VĂN TỐT NGHIỆP 70 LUẬN VĂN TỐT NGHIỆP 71 Fault Detection and Diagnosis Solution for PLC-Based Industrial Process Control Systems Dinh-Nhon Truong1, Huynh-The Pham2, Phuong-Nam Tran3, Van-Phuong Ta1,* HCMC University of Technology and Education, Thu Duc City, Ho Chi Minh City, Vietnam Cao Thang Technical College, Ho Chi Minh City, Viet Nam Ho Chi Minh City Vocational College, Ho Chi Minh City, Viet Nam * Corresponding author: phuongtv@hcmute.edu.vn 1Abstract—Industrial Process Control Systems (IPCS) play a key role in manufacturing plants There are many demands that need to tackle in the IPCS such as stability, robustness, process monitoring, troubleshooting, process control, effectiveness, and downtime In the above requirements, troubleshooting is considered one of the most important terms Consequently, this work proposed a fault detection and diagnosis (FDD) solution using the convolution neural network (CNN) for the PLC-based Industrial Process Control Model (IPCM) In particular, an S71500 PLC-based proportional integral derivative (PID) controller was utilized to control the IPCM to achieve stability and a training and diagnosis model was built by the 1DConvolution Neural Network (1D-CNN) to detect faults of sensors, actuators, signal modules, and controllers in the IPCM during operation The proposed solution not only obtains the desirable stability of the IPCM but also can detect, diagnose and alert specific errors that occur during the operation of IPCM on a Human Interface Machine (HMI) or a Personal Computer (PC) Through warning and error messages, operators or technicians can identify and troubleshoot faults quickly Index Terms— Fault Detection and Diagnosis (FDD); Programmable Logic Controller (PLC); Machine learning; Convolution Neural Network (CNN) INTRODUCTION Programmable Logic Controller (PLC)-based Industrial Process Control Systems (IPCS) play a crucial role in the industry field such as water and wastewater control systems, control and monitoring of plants, manufacturing control systems, chemical, petrochemical, and food industries [1] The IPCS usually has many sensors, actuators, modules, and controllers installed in the plants to meet the specific demands of manufacturing processes Detecting, diagnosing, and troubleshooting faults for the IPCS with a huge amount of devices is a very difficult task A mong device used in the IPCS, the central controller has the most important role It is considered the heart of the plant that controls, supervises, and diagnoses all the operations of the system Consequently, the central controller must have the ability to operate stably, robustly, and continuously in the industrial environment One of the industrial controllers that meet these requirements is the Programmable Logic Controller (PLC) [2]-[5] According to control algorithms, the PID controller is one of the most popular selections due to its simple and reliable structure [6] Therefore the PLCbased Proportional Integral Derivative (PID) controller has been developed, applied, and achieved impressive results in many areas including the industrial field [7]-[8] From the point of view of the manufacturing process, minimizing system downtime is crucial In order to prevent downtime in the plants, abnormal conditions or faults that happened during the operation need to detect, diagnose, and troubleshoot at the early stage Faulty causes of the plants can come from the sensors, actuators, modules, communication processes, and controllers Finding reasons and sources of faults exactly for the control systems, especially the industrial processes has been an urgent demand and attracted the attention of many researchers in recent years [9]-[11] For the latest PLCs, detection and diagnosis functions are integrated into signal modules and controllers to recognize hardware and software faults [12] Common faults such as wire breaks, no power supply, over range, and under range occur, and the codes automatically generate in specific blocks From the data generated in these blocks, Supervisory Control And Data Acquisition (SCADA) software can be used to manipulate, process and display corresponding fault messages [13-[14] Fig shows diagnosis fault blocks that are supported by the S7-1500 PLC Minor faults, however, such as signal attenuation on transmission lines, bad contacts at terminals, and bias and drift faults, the PLCs can not recognize and signal warning messages Therefore, the operators or engineers can not identify and fix these faults at the early stage As a result, the system will soon get major faults and downtime Artificial Neural Networks (ANNs) can mimic the brain of humans in the practice, training, and cognition process [15][16] Consequently, the ANNs have been developed and applied in learning, diagnosing, and predicting in various areas by many researchers [17]-[19] The basic structure of the ANN includes an input layer, a hidden layer, and an output layer Fig Diagnosis fault blocks supported in Siemens PLCs As the number of input and output data is massive, the mapping between input and output is non-linear and complex, the ANN needs more hidden layers that are called Deep Neural Networks (DNN) or deep learning (DL) to achieve realistic demands for learning, analysis, diagnosis, and prediction applications as shown in Fig [20]-[21] With the same intractable problems, the DL including more hidden layers will achieve higher accuracy [22]-[23] Hidden Input Output Layers Layer Layer Outputs Inputs F1 F2 F2 Fig A deep neural network structure In the deep learning architecture, the Convolutional Neural Network (CNN) is one of the most popular algorithms [24][26] A CNN is the feed-forward Artificial Neural Network (ANN) by adding the convolution and subsampling layers between input layers and hidden layers In particular, the CNN includes the input layer, convolution layers, filter channels, kernel matrices, pooling layers, flatten layer, fully connected layer, hidden layer, and output layer as shown in Fig By adding the convolution and subsample layers into the ANN structure, the CNN has feature extraction, learning, and diagnosis abilities more powerfully and accurately [27]-[31] Therefore, the CNN has been developed and applied in various fields including product classification, voice identification, fault detection, and prediction for commercial and industrial fields [32]-[40] Although, the CNN-based detection and prediction models have achieved good results in various areas, applying the CNN-based detection and diagnosis solution for PLC-based industrial process control systems has not been many publications yet This research paper proposed a detection and diagnosis solution for the PLC-based Industrial Process Control Model (IPCM) using 1D-CNN In particular, a PLC-based PID controller was utilized to control the IPCM to achieve desirable stability and a faulty learning and prediction model was built by the 1D-CNN for detecting and diagnosing faults of the IPCM during the operation With the proposed solution, faults that happened during the operation of the IPCM are early detected, diagnosed, and displayed alarms on the HMI or PC Through the alarm and warning messages, the operators or engineers can identify and troubleshoot the faults quickly The rest of this paper is organized as follows Section II describes the structure of an industrial process control model and intractable problems The fault detection and diagnosis solution for the PLC-based IPCM using 1D-CNN are proposed in Section III Section IV shows experimental results and evaluation The conclusions and next research are followed in Section V INDUSTRIAL PROCESS CONTROL MODEL A process control model produced by the INFINIT Inputs Filter Conv (2, ) Stride=(1, ) Flatten Convolution and Subsampling Layers AveragePooling(2) Stride=1 Filter Conv (2, ) Stride=(1, ) Hidden Layers AveragePooling(2) Stride=1 Outputs F1 F2 5x1 4x1x6 3x1x6 3x1x6 F2 2x1x3 6x1 Fig Structure of a Convolution Neural Network Fig Industrial Process Control Model Technologies are shown in Fig Similarly to the IPCS, the process control model has also many sensors and actuators including water level, pressure, and temperature sensors installed in the cylindrical glass tank, and the water pump and motorized valve are installed along with the pipe system to acquire data and control the system The measurable range of sensors and control signals of actuators are usually different from the acceptable signals of controllers Consequently, there is usually an interface unit to condition the signal between sensors, actuators and controllers Table I shows the parameters of sensors, actuators, and the output signal of the interface unit In order to meet the industrial environment, an S7-1500 PLC was used to control the model in this research In the non-fault condition, the sensor signal range, control signal and output signal of the interface unit are shown Table However, the output signal of the interface unit will be diverted from the standard range as soon as the faults occur Table II illustrates the fault cases and output corresponding signals of the interface unit According to the output signals of the interface unit in Table I and Table II, there are some serious faults (from F6 to F10), and the output signals have a large difference from the faultless case (F0) On the other way, for some minor faults (from F1 to F5), the output signals not have much difference from the faultless case Therefore, the fault detection and diagnosis solution based on only the diagnosis3 fault blocks of the PLC can not cover all fault cases As a result, the system will soon get major faults and downtime in case faults can not be detected and recovered in a timely manner TABLE I: RANGE OF MEASUREMENT AND CONTROL SIGNALS IN FAULTLESS CASE No Devices Range of Standardized measurement electrical signal by the interface unit Level 5cm to 13 cm to 10 volts sensor Flow to 7,5 litres to 10 volts sensor per minute Pressure to bar to 10 volts sensor Water to 7,5 litres to 10 volts pump per minute TABLE II: FAULT CAUSES AND OUTPUT SIGNALS Fault Causes The output signal codes of the interface unit F1 The water level From to volts sensor signal attenuation was from 10% to 15% F2 The water level From to 8.5 volts sensor signal attenuation was over 15% F3 Water flow sensor From to volts signal attenuation was from 10% to 15% F4 Water flow sensor From to 8.5 volts signal attenuation was over 15% F5 Pressure sensor From to volts signal attenuation was from 10% to 15% F6 Pressure sensor From to 8.5 volts signal attenuation was over 15% F7 The water level volt sensor signal was lost F8 The water flow volt sensor signal was lost F9 The pressure volt sensor signal was lost F10 The pump control volt signal was lost FAULT DETECTION AND DIAGNOSIS SOLUTION FOR THE PLC-BASED INDUSTRIAL PROCESS CONTROL MODEL USING 1D-CNN The industrial process control model has many parameters that need to qualify during the operation to assess whether the system is in faulty conditions As any sensor or actuator gets fault, the output and control signals will have a significant deviation from the standard range Besides, it also affects the data of other devices and vice versa The data distribution of the water level sensor, pressure sensor, flow sensor, and control signal in faultless and faulty cases are shown in Fig 5, Fig 6, Fig 7, and Fig 8, respectively In order to detect and diagnose all faulty cases of sensors and actuators, a fault detection and diagnosis solution for the PLC-based process control model using 1D-CNN was proposed in this research A block diagram of the proposed solution is shown in Fig The proposed fault detection and diagnosis solution include the industrial process control model installed sensors, actuators, and the interface signal unit for measuring, controlling, and signal condition; the industrial controller S71500 PLC is used to control the system to achieve the desired performance; a PC installed Python for data access from S71500 PLC in real-time via Github library; a learn and diagnosis model was built by 1D-CNN using Python to predict faults during the operation of the process control model The procedures for reading and writing data between Python and S7-1500 PLC are shown in Fig 10 The IPCM has five variables that need to qualify during the operation including set point, process value, control signal, flow, and pressure The IPCM is a class of time-driven data Therefore, 1D-CNN was chosen to build the learn and diagnosis model for prediction faults during the operation The structure of the learning and diagnosis model for the IPCM using 1D-CNN is shown in Fig 11 According to the 1D-CNN model, the number of convolution, subsampling, hidden layers, and neurons in each layer affects the accuracy of the learning and prediction process There are no specific principles to config exactly the numbers of hidden layers and neurons in convolution neural networks Furthermore, the performances of learning algorithms also depend on the structure and the characteristics of the system A learning algorithm which has good performance with this system can not make sure to have effectiveness for the other systems The structure selection of the learning and prediction model usually obtains based on references and practical experiments For this work, the learning and diagnosis model for the IPCM using 1D-CNN includes two convolution layers followed by two average spooling, one flattene layer, two hidden neural layers, and one 11-neural output layer to predict the normal and faulty cases during the operation of the system Fig Data distribution of water level sensor with fault and non-fault cases Fig Data distribution of flow sensor with fault and non-fault cases Fig.7 Data distribution of pressure sensor with fault and non-fault cases Fig Data distribution of control signal with fault and non-fault cases PC Industrial Controller Industrial Process Control Dislay Types of Faults Select prediction model Build Model for Training and Evaluation Acquire dataset in normal and fault modes Acquire datasets in in real-time Interface Unit Fault Detection, Diagnosis, alarm, and Control Interface Read sensor signalsUnit Actuator Control signals Fig The block diagram of the FDD Fig 10 The procedures for reading and writing data between Python and S7-1500 PLC Sensor Signals Time Each segment includes 20 samples, each sample has features Setpoint Each segment includes 20 samples, each sample has features Each segment includes 20 samples, each sample has features Process Value Control Signal Flow Pressure Filter Conv (5, ) Stride=(1, ) AveragePooling(2) Stride=2 20x1X6 Filter 16 Conv (5, ) Stride=(1, ) AveragePooling(2) Stride=2 10x1x6 10x1x16 5x1x16 Output Full Connected Layers Flatten() Normal F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 80 84 120 11 Fig 11 1D-CNN model for the IPCM I.EXPERIMENTAL RESULTS AND EVALUATION In order to assess the operation of the proposed solution in real-time, the PLC-PID controller was used to control the IPCM to achieve the desired set points The parameters of the PID controller were achieved by the auto-turning function in S7 1500 PLC with Kp = 0.41893, Ki = 0.64842, and Kd = 0.13275 The tracking response of the water level in the cylindrical glass tank is shown in Fig 12 Along with obtaining the desired set points, the 1D-CNN in Fig 11 was utilized to learn and diagnose faults of the IPCM in real-time In particular, the datasets containing 17600 samples were divided into segments Each segment contains 20 samples and each sample has features including set point, process value, control signal, flow, and pressure The batch size and epoch are 32 and 1000 respectively The total fault and faultless cases at the outputs are 11 With the above 1D-CNN configuration, the learning time of the model was 836 seconds and 887.43 seconds for the 1DCNN with sample and segment inputs respectively For accuracy, the 1D-CNN with segment input obtained 99.88% in comparison with 99.51% of the 1D-CNN with sample input Table summarizes the parameters of the learning and prediction model for the IPCM using 1D-CNN in practice According to the fault diagnosis solutions using the convolution neural network, there have been many publications in recent years In particular, L.Eren et al proposed fault diagnosis and classifier methods with an accuracy of 93,9%, 97,8%, 98%, and 99%, respectively [32][35] With novel 1D-Convolution neural networks, the performance has been improved significantly In [36], [37],[38], and [39], the accuracy of fault diagnosis solutions achieved at 99.2%, 99.7%, 98.5%, and 99.75% In comparison with the recent research, the proposed solution in this work achieved an impressive performance at 99.88% for the PLC-based industrial process control model in real-time Along with predicting faults exactly, the proposed solution also generated automatically specific warning and error messages on the PC or HMI TABLE 3: COMPARISON BETWEEN 1D-CNN SAMPLE AND SEGMENT INPUTS Algorithms 1D-CNN 1D-CNN with sample with segment Parameters input input Total samples Number of healthy samples Number of faulty samples Number of features Number of outputs CNN input Batch size Epoch Execution Time in second Accuracy 17600 1600 17600 1600 16000 16000 11 sample 32 1000 887.43 11 segment 32 1000 836.36 99.51% 99.88% Through the message system, the operators or engineers can identify the reasons and the source of the faults Consequently, they can troubleshoot the system quickly to keep system from downtime Fig 13 shows the error and warning messages of faults every second during the operation of the IPCM in real-time Fig 12 Tracking response of the water level in the tank Fig 13 Warning and error message system II.CONCLUSION AND NEXT WORKS In this research, the water level in the cylindrical glass tank of the IPCM was controlled by the PLC-based PID controller to achieve the desired performance in real time In addition to obtaining the stability of the water level, different faults that happened during the operation of the system were also detected, and diagnosed by the 1D-CNN with high accuracies The diagnosis results were displayed on the PC or HMI to show specific faults Through alarm and warning messages and fault codes, the operators or engineers can troubleshoot issues quickly With the proposed solution, the troubleshooting for industrial process control systems with a huge amount of devices will be handled exactly and timely PLC-based industrial process control systems have become very popular in manufacturing plants Therefore, achieved results from this research can adapt to realistic requirements Although the proposed solution has achieved good results in learning and diagnosing different faults of the industrial process control model in real-time, applying it to manufacturing plants should be considered in many aspects such as simulation fault methods during the operation of the system to acquire datasets, downtime for building learn and prediction model, integration of fault detection and diagnosis solution into the existing program of the system ACKNOWLEDGEMENT This research is supported by Ho Chi Minh City University of Technology and Education (HCMUTE), Vietnam CONFLICTS OF INTEREST We declare that they have no conflicts of interest to report regarding the present study REFERENCES Ephrem RyanAlphonsus and Mohammad Omar Abdullah, “A review on the applications of programmable logic controllers (PLCs)”, Renewable and Sustainable Energy Reviews, vol 60, pp 1185-1205, 2016 DOI: doi.org/10.1016/j.rser.2016.01.025 [2] Zafer Aydogmus, “A Real-Time Robust Fuzzy-based Level Control Using Programmable Logic Controller”, Elektronika ir Elektrotechnika, vol 21, no.1, pp 13-17, 2015 DOI: 10.5755/j01.eee.21.1.7812 [3] G V Palomo and J A Rossiter, “Programmable logic controller implementation of an auto-tuned predictive control based on minimal plant information”, Elsevier, vol 50, no 1, pp 92-100, 2011 DOI: 10.1016/j.isatra.2010.10.002 [4] C Parrott and G K Venayagamoorthy, “Real-time implementation of intelligent modelling and control techniques on a PLC platform”, IEEE Industry Applications Society, Edmonton, AB, Canada, pp 1-7, 2008 DOI: 10.1109/08IAS.2008.164 [5] Van-Phuong Ta*, Dinh-Nhon Truong, Nguyen-Thanh Nhan, “An Innovative Approach for Water Distribution Systems”, Intelligent Automation & Soft Computing, vol.32, no 3, pp 1605-1615, 2022 DOI:10.32604/iasc.2022.022374 [6] Jiangbo Z, Junzheng W, “The fractional order PI control for an energy-saving electro-hydraulic system” Trans Inst Meas Control vol 39, pp 505-519, 2015 DOI: 10.1177/0142331215610 [7] Rakesh P Borase, D K Maghade, S Y Sondkar1 and S N Pawar, “A review of PID control, tuning methods and applications”, International Journal of Dynamics and Control, vol 9, pp 818-827, 2020 DOI: 10.1007/s40435-020-00665-4 [8] Kagan Koray Ayten, Ahmet Dumlu, “Implementation of a PID Type Sliding-Mode Controller Design Based on Fractional Order Calculus for Industrial Process System”, Elektronika ir Elektrotechnika, vol 27, no.6, pp 4-10, 2021 DOI: 10.5755/j02.eie.30306 [9] Wagner Fontes Godoy, Ivan Nunes da Silva , Alessandro Goedtel, Rodrigo Henrique Cunha Palácios, “Evaluation of stator winding faults severity in inverter-fed induction motors”, Applied Soft Computing, vol 32, pp 420–431, 2015 DOI: 10.1016/j.asoc.2015.03.053 [10] Wagner F Godoy, Ivan N da Silva, Alessandro Goedtel, Rodrigo H C Palacios, Gustavo H Bazan, and Daniel ´ Mor´ ınigo-Sotelo, “An application of artificial neural networks and PCA for stator fault diagnosis in inverter-fed induction motors”, XXII International Conference on Electrical Machines (ICEM), pp 21652171, 2016 DOI: 10.1109/ICELMACH.2016.7732822 [11] Debashish Mohapatra, Bidyadhar Subudhi, Raju Daniel, “Real-time sensor fault detection in Tokamak using different machine learning algorithms”, Fusion Engineering and Design, vol 151, pp.18, 2020 DOI: 10.1016/j.fusengdes.2019.111401 [12] Michael Steinegger, Martin Melik-Merkumians, Johannes Zajc, and Georg Schitter, “Automatic generation of diagnostic handling code for decentralized PLC-based control architectures”, IEEE 21st International Conference on Emerging Technologies and Factory Automation (ETFA), pp.1-8, 2016 [1] DOI: 10.1109/ETFA.2016.7733694 [13] Jose Miguel Gutierrez-Guerrero, Juan Antonio Holgado-Terriza “iMMAS an Industrial Meta-Model for Automation System Using OPC UA”, Elektronika ir Elektrotechnika, vol 23, no.3, pp 3-11, 2017 DOI: 10.5755/j01.eie.23.3.18324 [14] A Venuprasad and Y S Narayan, “Automatic monitoring and controlling of pressure using plc and SCADA,” International Journal of Innovative Research in Science, Engineering and Technology, vol 5, no 7, pp 13671-13677, 2016 DOI:10.15680/IJIRSET.2016.0507225 [15] AboozarTaherkhani, AmmarBelatreche, YuhuaLi, GeorginaCosma, Liam P.Maguire, T.M.McGinnity, “A review of learning in biologically plausible spiking neural networks”, Neural Networks, vol 122, pp.253-272, 2020 DOI: 10.1016/j.neunet.2019.09.036 [16] YihaoWu, MasahikoMorita, JunIzawa, “Reward prediction errors, not sensory prediction errors, play a major role in model selection in human reinforcement learning”, Neural networks, vol.154, pp.109121, 2022 DOI: 10.1016/j.neunet.2022.07.002 [17] Omar Aguilar-Mejía, Rubén Tapia-Olvera, Antonio ValderrabanoGonzález, and Iván Rivas Cambero “Adaptive neural network control of chaos in permanent magnet synchronous motor”, Intelligent Automation and Soft Computing, Vol.22, no.3, pp.1–9, 2016 DOI: 10.1080/10798587.2015.1103971 [18] Javier Moreno-Valenzuela, Carlos Aguilar-Avelar, Sergio PugaGuzmán & Víctor Santibáđez, “Two adaptive control strategies for trajectory tracking of the inertia wheel pendulum: neural networks vis vis model regressor”, Intelligent Automation and Soft Computing, vol 23, No 1, pp 1-11, 2017 DOI: 10.1080/10798587.2015.1121618 [19] S Jung, “Stability Analysis of Reference Compensation Technique for Controlling Robot Manipulators by Neural Network”, International Journal of Control, Automation and Systems, vol 15, no 2, 2017, pp 952-958 DOI: 10.1007/s12555-015-0070-7 [20] Radoslaw M Cichy and Daniel Kaiser, “Deep Neural Networks as Scientific Models”, Trend in Cognitive Sciences, vol 23, No 4, 2019 DOI: 10.1016/j.tics.2019.01.009 [21] Feng Jia, Yaguo Lei, Jing Lin, Xin Zhou, Na Lu, “Deep neural networks: A promising tool for fault characteristic mining and intelligent diagnosis of rotating machinery with massive data”, Mechanical Systems and Signal Processing, vol 72, pp 303-315, 2016 DOI: 10.1016/j.ymssp.2015.10.025 [22] Heaton, J Ian Goodfellow, Joshua bengio, and aaron courville: “Deep learning”, Genet Program Evol Mac vol 19, pp 1-3, 2017 DOI: 10.1007/s10710-017-9314-z [23] Laith Alzubaidi, Jinglan Zhang, Amjad J Humaidi, Ayad AlDujaili, Ye Duan, Omran Al-Shamma, J Santamaría, Mohammed A Fadhel, Amidei & Laith Farhan, “Review of deep learning: concepts, CNN architectures, challenges, applications, future directions”, Journal of Big Data, vol 8, no 52, pp 1-74, 2021 DOI: 10.1186/s40537-021-00444-8 [24] Gu J, Wang Z, Kuen J, Ma L, Shahroudy A, Shuai B, Liu T, Wang X, Wang G, Cai J, et al, “Recent advances in convolutional neural networks”, Pattern Recognition, vo 77, pp 354-377, 2018 DOI: 10.1016/j.patcog.2017.10.013 [25] Serkan Kiranyaz , Onur Avci, Osama Abdeljaber, Turker Ince,Moncef Gabbouj, Daniel J Inman, “1D convolutional neural networks and applications: A survey”, Mechanical Systems and Signal Processing, Vol 151, pp 1-5, 2021 DOI: 10.1016/j.ymssp.2020.107398 [26] JinyangJiao, MingZhao, JingLin, KaixuanLiang, “A comprehensive review on convolutional neural network in machine fault diagnosis”, Neurocomputing, vol 417, no 5, pp 36-63, 2020 DOI: 10.1016/j.neucom.2020.07.088 [27] Ridvan Yayla, Baha Sen, “A New Classification Approach with Deep Mask R-CNN for Synthetic Aperture Radar Image Segmentation”, Elektronika ir Elektrotechnika, vol 26, no.6, pp 52-57, 2020 DOI: 10.5755/j01.eie.26.6.25849 [28] GuoqingLi, MengZhang, JiaojieLi, FengLv, GuodongTong, “Efficient densely connected convolutional neural networks, Pattern recognition”, vol 109, pp 1-9, 2021 DOI: 10.1016/j.patcog.2020.107610 [29] Hongmei Li, Jinying Huang and Shuwei Ji, “Bearing Fault Diagnosis with a Feature Fusion Method Based on an Ensemble Convolutional Neural Network and Deep Neural Network”, Sensors, vol.19, no.9, pp.1-18, 2019 DOI: 10.3390/s19092034 [30] Fikret Yalcinkaya, Ali Erbas, “Convolutional Neural Network and Fuzzy Logic-based Hybrid Melanoma Diagnosis System”, Elektronika ir Elektrotechnika, vol.27, no 2, pp 69-77, 2021 DOI: 10.5755/j02.eie.28843 [31] O Abdeljaber, O Avci, S Kiranyaz, M Gabbouj, D.J Inman, “Realtime vibration-based structural damage detection using onedimensional convolutional neural networks”, J Sound Vib, vol 388 2017 DOI: 10.1016/j.jsv.2016.10.043 [32] L Eren, T Ince, S Kiranyaz, “A generic intelligent bearing fault diagnosis system using compact adaptive 1D CNN classifier”, J Signal Process Syst Vol 91, pp 179–189, 2019 DOI: 10.1007/s11265-018-1378-3 [33] T Ince, S Kiranyaz, L Eren, M Askar, M Gabbouj, “Real-time motor fault detection by 1-D convolutional neural networks”, IEEE Trans Ind Electron Vol 63, pp 7067–7075, 2016 DOI: 10.1109/TIE.2016.2582729 [34] Zhuyun Chen, Alexandre Mauricio, Weihua Li, Konstantinos Gryllias, “A deep learning method for bearing fault diagnosis based on Cyclic Spectral Coherence and Convolutional Neural Networks”, Mechanical Systems and Signal Processing, vol 40, 2020 DOI: 10.1016/j.ymssp.2020.106683 [35] O Abdeljaber, S Sassi, O Avci, S Kiranyaz, I Abulrahman, M Gabbouj, “Fault detection and severity identification of ball bearings by online condition monitoring”, IEEE Trans Ind Electron, 2018 DOI: 10.1109/TIE.2018.2886789 [36] Chih-Cheng Chen, Zhen Liu, Guangsong Yang, Chia-Chun Wu, and Qiubo Ye, “An Improved Fault Diagnosis Using 1D-Convolutional Neural Network Model”, Electronics, vol 10, no 59, 2021 DOI: 10.3390/electronics10010059 [37] S Kiranyaz, A Gastli, L Ben-Brahim, N Alemadi, M Gabbouj, “Real-time fault detection and identification for MMC using 1D convolutional neural networks”, IEEE Trans Ind Electron, 2018 DOI: 10.1109/TIE.2018.2833045 [38] Huahui Yang, Chen Meng, Cheng Wang, “Data-Driven Feature Extraction for Analog Circuit Fault Diagnosis Using 1-D Convolutional Neural Network”, ”, IEEE Access, vol 8, pp 1-9, 2020 DOI: 10.1109/ACCESS.2020.2968744 [39] FengXue, WeiminZhang, FeiXue, DongdongLi, ShulianXie, JürgenFleischer, “A novel intelligent fault diagnosis method of rolling bearing based on two-stream feature fusion convolutional neural network”, Measurement, vol 176, pp 1-19, 2021 DOI: 10.1016/j.measurement.2021.109226 [40] Shengnan Tang, Shouqi Yuan and Yong Zhu, “Convolutional Neural Network in Intelligent Fault Diagnosis Toward Rotatory Machinery”, IEEE Access, vol 8, pp 86510 – 86519, 2020 DOI: 10.1109/ACCESS.2020.2992692 S K L 0 ... lượng nước để kiểm tra mức độ ổn định bơm ❖ Thiết kế mạch giả lập lỗi cho hệ thống 3.1.2 Yêu cầu phần mềm Đối với đề tài ? ?Ứng Dụng Deep Learning Để Nhận Dạng Lỗi Trong Hệ Thống Cơng Nghiệp? ?? có... Điện tử Tên đồ án, luận án môn thi tốt nghiệp: Ứng Dụng Deep Learning Để Nhận Dạng Lỗi Trong Hệ Thống Công Nghiệp Ngày & nơi bảo vệ đồ án, luận án thi tốt nghiệp: 06/11/2022 Trường ĐH Sư Phạm Kỹ... CƠ SỞ LÝ THUYẾT 2.1 HỆ THỐNG ĐIỀU KHIỂN CÔNG NGHIỆP 2.1.1 Thành phần hệ thống điều khiển công nghiệp .6 2.1.2 Ứng dụng hệ thống điều khiển công nghiệp thực tế 2.1.3 Những