Artificial intelligent fault diagnosis system of complex electronic equipment

Abstract

The traditional fault detection system of complex electronic equipment based on image analysis theory only analyzes the image characteristics of complex electronic equipment for artificial intelligent fault diagnosis. It cannot deal with the system diagnosis problem of qualitative fault data and has the problems of low accuracy and long time consuming of fault detection. To address these problems, an artificial intelligent fault diagnosis system of complex electronic equipment based on BP neural network is designed in this paper. BP neural network model for artificial intelligent fault diagnosis of complex electronic equipment is built based on system overall structure. The structure of BP neural network and learning algorithm is determined according to the actual fault problem. Learning and training of BP neural network are carried out by using sample data of fault. Artificial intelligent fault diagnosis algorithm of complex electronic equipment based on BP neural network and qualitative fault data is used, which combines the BP neural network and qualitative fault data. The preprocessing method is applied to quantify the fault data. Fault diagnosis is achieved by BP neural network technology. The system database and the implementation process of the BP neural network are designed. Experimental results show that the designed system can significantly improve the accuracy of fault detection of complex electronic equipment, improve the effect of fault detection, and reduce the time consuming of fault detection.

Keywords

Complex electronics equipment fault artificial intelligence diagnostic system BP neural network residual signal

1 Introduction

In the process of long time storage and protection of large and complex equipment, the performance indexes of large and complex equipment will appear to be decayed with various factors, and there are various faults. Fast and accurate fault diagnosis is an important part of improving the combat effectiveness of complex electronic equipment. Some of the equipment is integrated system of light, machine and electricity. The relationship between the components and the subsystems is complex. The working state and the environment are varied, which results in the high nonlinearity and obvious characteristics of these fault data [1]. For example, failure data is difficult to be quantified, which is not conducive to quantitative processing. The occurrence of some faults is few and suddenly. The data collection is difficult and the size of sample data is small. Fault is affected by many factors. The data uncertainty is strong and the mapping relationship between fault location and fault localization does not have one-to-one correspondence. The traditional fault detection system of complex electronic equipment based on image analysis theory only analyzes the image characteristics of complex electronic equipment for artificial intelligent fault diagnosis. It cannot deal with the system diagnosis problem of qualitative fault data, which results in the problems of low accuracy and long time consuming of fault detection [2]. To address these problems, an artificial intelligent fault diagnosis system of complex electronic equipment based on BP neural network is designed in this paper. The designed system can improve the accuracy of fault detection with artificial intelligence and shorten the time consuming of fault detection.

2 Artificial intelligent fault diagnosis system of complex electronic equipment based on BP neural network

2.1 System structure

The overall structure of intelligent fault diagnosis system of complex electronic equipment is shown in Fig. 1. The detected object is the actual equipment and simulation equipment of a part of the equipment. The collected signal is sent to the main computer by serial communication. The functions of signal processing and fault diagnosis are implemented in computer software system. The software system consists of 4 functional modules, which are the user management module, the database module, the human-computer interaction module and the intelligent diagnosis module [3]. The core of the system is the intelligent diagnosis module based on BP neural network. In this paper, the intelligent fault diagnosis module based on BP neural network is implemented on the platform of VC++ 6.0 and the design of low level database with MicrosoftAccess2003.

Fig.1

Overall structure of intelligent fault diagnosis system of complex electronic equipment.

2.2 Building of BP neural network model for artificial intelligent fault diagnosis system of complex electronic equipment

Based on the overall structure of system, a comprehensive analysis of the complex relationship between multiple fault causes and multiple fault symptoms for complex electronic equipment is carried out. Then BP neural network model for artificial intelligent fault diagnosis system of complex electronic equipment is built.

2.2.1 BP neural network

In theory, the three-layer BP neural network can approximate any continuous function with arbitrary precision when the number of neurons in the hidden layer is enough. The model is shown in Fig. 2.

Fig.2

BP neural network model of fault diagnosis.

x_i is the input of the i_th neuron in the input layer, which represents the i_th fault symptom in the fault mode. y_i is the output of the i_th neuron in the output layer, which represents the i_th fault cause in the fault mode. $w_{ij}^{1}$ is the connection weight of the i_th neuron of the (l - 1) _th layer and the j_th neuron of the l_th layer (the hidden layer l = 1, the outputlayer l = 2).

The BP neural network method is to take the square sum of the difference of the actual output value and the desired value as the objective function of on the premise of the given output expectation [4]. The objective function is minimized by the adjustment of the weights and the thresholds. It mainly includes two parts: forward propagation and error reverse correction. Forward propagation means that the input of each layer of node only comes from the output of the previous layer of node. The output O₁ of the neuron in the input layer is equal to the input X_i. The output of a neuron in the hidden or output layer is givenby $O_{j}^{l} = f (\sum_{i = 1} w_{ij}^{l} o_{i}^{l - 1} + θ_{j}^{l}$ (1) where f is the transform function, here taking the sigmiod function, $θ_{j}^{l}$ is the neuron threshold of the l_th layer, regarded as the connection weight of the current neuron and a virtual neuron in the l = 1 layer [5]. If the desired output of the j_th neuron in output layer is T_j, total output error of output layer (objective function) is given by $E = \frac{1}{n} \sum_{j = 1}^{n} [T_{j} - O_{2 j}]^{2}$ (2)

The gradient descent method is used to solve the minimum value of the objective function, and the smoothing factor β is introduced to improve the convergence speed. Error back propagation is to adjust the connection weight of each node from the output layer along the gradient of the objective function according to output error, so that the objective function is gradually reduced to a certain threshold (set to 0.5 in this paper) or no longer to reduce [6]. The adjustment of network weight and node threshold formula is given by

$\begin{matrix} w_{ij}^{l} (k + 1) & = & w_{ij}^{l} (k) + η δ_{j}^{l} O_{i}^{l - 1} \\ + β (w_{ij}^{l} (k) - w_{ij}^{l} (k - 1)) \end{matrix}$ (3) where K is the number of iterations, η is the Gain coefficient (learning rate), and $δ_{j}^{1}$ is the back propagation error signal, the values in the output layer and the hidden layer are given by $\begin{matrix} δ_{j}^{2} = f^{'} (\sum_{i = 1} w_{ij} o_{i} + θ_{j}) \\ (\sum_{i = 1} w_{ij} o_{j} - f (\sum_{i = 1} w_{ij} o_{i} + θ_{j})) \end{matrix}$ (4) $δ_{j}^{1} = f^{'} (\sum_{i = 1} w_{ij} o_{j} + θ_{j} \sum_{n} δ_{j}^{2} w_{ij}^{2}$ (5)

2.2.2 Learning training and fault diagnosis process of BP neural network

In the software of the system, human-computer interaction is used to achieve the fault diagnosis of complex electronic equipment. The software includes the functions of adding the feature data of equipment samples, selecting feature data of equipment samples, building BP neural network and training network, saving successfully trained BP neural network, input fault data to be diagnosed [7], and fault diagnosis and output results with the trained BP neural network. Figure 3 shows the overall structure flow chart of the software system.

Fig.3

The overall structure flow chart of the software system.

Because different complex electronic equipment may has different characteristics and output information, that is, that is, different number of neurons in the input layer and output layers, the universality of software is considered in design, and the module of adding feature data is added. If other equipment needs to be diagnosed, special diagnostic data is added with this module, so that software can achieve fault diagnosis for multi-equipment [8]. The software has also designed the module of selecting sample feature data. The purpose is that the same type of equipment does not need to repeat adding the feature data in the fault diagnosis, and directly selects the existing feature data module. In addition, considering the parameters that need to be set up in training network, such as number of hidden layer neurons, learning efficiency, training times, and training objective may be different for different networks, some parameters of the system software can bemodified [9].

The flow chart of learning training and fault diagnosis of the BP neural network model is shown in Fig. 4. Firstly, the structure and learning algorithm of BP network are determined according to the actual problems, and then the network is trained by the sample data of several groups of fault symptoms and causes. The steps of training are as follows.

Network weights and node thresholds are initialized as a set of random numbers.

Input P groups of training sample. Input a set of fault symptom sample $X^{P} = {X_{1}^{P}, X_{2}^{P}, . . . X_{m}^{P}}$ . Set the desired output $T^{P} = {t_{1}^{P}, t_{2}^{P}, . . . t_{n}^{P}}$ .

According to BP network, the error of $Y^{P} = {y_{1}^{P}, y_{2}^{P}, . . ., y_{n}^{P}}$ and $T^{P} = {t_{1}^{P}, t_{2}^{P}, . . . t_{n}^{P}}$ is calculated to modify adaptively the connection weight and node threshold for minimizing the objective function.

The training samples are selected in turn, and the above process [10] is carried out continuously, until the total error of P groups of samples meets the expected requirement.

Fig.4

BP neural network fault diagnosis system flow chart.

Fig.5

Training precision of BP network based on qualitative data.

The trained fault knowledge model is stored in the BP network knowledge base. When diagnosing, fault symptom vector is used as input of network. The output result of network after calculating is compared with network knowledge base to find out the fault knowledge model which is closest to the sample to be diagnosed. In this way, the fault diagnosis isrealized.

2.3 Artificial intelligent fault diagnosis algorithm of complex electronic equipment based on BP neural network and qualitative fault data

The artificial intelligent fault diagnosis algorithm of complex electronic equipment based on BP neural network and qualitative fault data is to quantify the qualitative description of the failure phenomenon by using a rough set based fault data preprocessing method [11]. Then, the BP neural network technology is used for fault diagnosis. The preprocessed fault data can effectively enhance the identification and reduce the fault detection time and training steps. The modeling of qualitative data is as follows.

Step 1. Preprocessing of qualitative fault data and design of network input and output

The BP network is a multi-input single-output model, and the output range is [0,1]. However, each fault of the equipment may be mapped to many reasons, that is, the result of diagnosis is not unique, and it is necessary to determine the true cause of the fault by manual investigation. In order to apply to the BP network, the number of the fault phenomena is adjusted as follows.

The fault phenomena is numbered with 4-digit number and used as an input, and the input dimension is 1.

There are many causes for the fault, while the BP output is single. So the cause of the fault is used as follows: the number of the causes of a fault phenomenon is taken as the output [12]. For example, if there are three causes of the 1101 fault, the output of the BP network is set to 3. And the number of fault causes will be represented by binary numbers.

From the collected data, the number of fault cause is up to six. As the number of causes may be increased, a binary code with more than three digits is used. On the one hand, it is for supplement of fault cause, and data change is simple [13]. On the other hand, because of the little difference between data, it is easier to use the normalized preprocessing, and meet the output requirements of BP model. Table 1 shows a 3-digit binary encoding table for the number of some fault causes. The diagnosis result is a single output of a three-dimensional row vector.

Table 1
3-digit binary encoding table for the number of fault causes

Serial number Number causes Binary coding

1 0 [000]

2 1 [001]

3 2 [010]

4 3 [011]

5 4 [100]

6 5 [101]

7 6 [110]

8 7 [111]

8 7 [111]

Serial number	Number causes	Binary coding
1	0	[000]
2	1	[001]
3	2	[010]
4	3	[011]
5	4	[100]
6	5	[101]
7	6	[110]
8	7	[111]
8	7	[111]

Step 2. Building of BP neural network

The simulation experiment is carried out by Matlab. The Matlab neural network toolbox NNET provides a rich neural network realization function, including graphical user interface function [14], neural network creation, training and simulation function, drawing function, and Simulink support. The implementation of BP neural network with Matlab is obtained by using the following functions.

Newff (.): BP neural network creation function.

Train (.): network training function.

Sim (.): use the network for simulation.

The setup network has three layers: the input layer, the output layer, and the hidden layer. The coding of the fault phenomenon of certain equipment is taken as the input, then the number of input layer nodes n = 1. The output is a three-dimensional row vector that characterizing the number of fault causes, so the number of output nodes is m = 3. The implicit layer is used to approximate any rational function [15]. The number of nodes l is determined by the empirical formula l = 2n + a or $l = \sqrt{m + n} + a$ (a ∈ [1, 10]).

The value of a: The minimum mean square error is usually set to 1e-8, and the number of training times is 1000. But through simulation, it is found that no matter what empirical formula is used and what value a is, the performance of the trained BP network cannot reach the expected 1e-8, and the training steps to achieve stability are within 500. The training precision is shown in Table 2.

Table 2

Training precision

a	1	accuracy
1	3	0.162810
2	4	0.159797
3	5	0.162692
4	6	0.133587
5	7	0.141057
6	8	0.139918
7	9	0.125075
8	10	0.127517
9	11	0.119942
10	12	0.114898

Since the actual precision of 1000 times of training is between 1e-2 and 1e-1, and more than 0.11 and less than 0.17, the middle value can be taken and the training precision is 0.14. The simulation results show that the precision is about 0.14 and the training step is relatively less when the stability is reached [16]. At this time l = 7, a = 5. Thus the number of hidden nodes can be determined to be 7.

Step 3. Training of BP neural network

The BP network training results obtained with qualitative data are shown in Fig. 4, in which the solid line indicates the target accuracy and the dotted line indicates the training precision.

Step 4. BP network test and fault diagnosis

The qualitative sample data is used to test network. The test results are 98 groups, of which 81 groups are normal output and 7 groups are not consistent with target output. The simulation accuracy is 82.7%, which basically meets the expected requirements of fault diagnosis. This method needs to complete the diagnosis task together with other diagnostic methods for the new faults and the faults which are non-existent in database. After the diagnosis is completed, the database is supplemented in a unified coding form [17], and the neural network is retrained.

2.4 Implementation of BP neural network model for fault diagnosis of complex electronic equipment

2.4.1 Design of database

The database of the equipment intelligent fault diagnosis system consists of 9 tables, and the composition and functions are shown in Table 3.

Table 3
Composition and function table of the system database

Serial number Table name Function

1 DEVICE_TB Equipment element information

2 FAULT_TB Fault cause information

3 OMEN_TB Fault symptom information

4 SAMPLE_TB sample data

5 SAMPLE_TIB Temporary sample data

6 WHI_TB Hidden and input layer weights

7 WHI_TTB Implicit and input layer temporary weights

8 WHO_TB Output layer and hidden layer weight

9 WHO_TTB Temporary weights of the output layer and the hidden layer

Serial number	Table name	Function
1	DEVICE_TB	Equipment element information
2	FAULT_TB	Fault cause information
3	OMEN_TB	Fault symptom information
4	SAMPLE_TB	sample data
5	SAMPLE_TIB	Temporary sample data
6	WHI_TB	Hidden and input layer weights
7	WHI_TTB	Implicit and input layer temporary weights
8	WHO_TB	Output layer and hidden layer weight
9	WHO_TTB	Temporary weights of the output layer and the hidden layer

For the table of FAULT_TB, the fields in the table include the serial number of the fault cause (ID), the serial number of the fault component (Device ID), the fault description (Fault DES), and the maintenance strategy (Maintain). The data types of each field are set to automatically numbered, digital, and text, and serial number of the fault cause is the primary key.

2.4.2 Implementation of BP neural network model

First, fault symptom and fault cause are set on the human-computer interaction interface of the system, including setting the number of neurons in the input layer and the output layer m and n and the names of the input and the output. The set information is saved into the variables and the database, and FAULT_TB and OMEN_TB tables. The number of neurons in the hidden layer p is set to $p = \sqrt{m + n} + 2$ . Set the number of samples and input the sample data. In this paper, the learning training of BP network is mainly implemented by the function of Void CSampleLean:training (int inlayer int outLayer int Hide Layer int Size)

{…}

The steps of the implementation are as follows.

Define variables and connect the database [18]. Define dynamic array (for storage of connection weight, input value, and output value). The class CArray<TYPE provided by MFC, ARG_TYPE>provided by MFC is used to implements the operations of dynamic change of the array, addition and delete of array variables and data.

Initialize the connection weights with a random number.

Enter the sample learning cycle. The output error of a set of samples is calculated according to the formula in the paper and the weights are adjusted, then the sample of the next group is calculated until the learning of all the samples is completed.

Calculate total error of output. If the error reaches the expected requirement, go to the step (5). Otherwise, the difference between this iteration and the previous iteration is calculated. If it is greater than a certain value [19], go to the step (3) for reiteration. Otherwise, go to the step (5).

Finish the learning training and save the final data of connection weight into the database.

3 Experimental analysis

Experiments are carried out to verify the effectiveness of the fault diagnosis system. Taking a component of complex electronic equipment as an example, the fault symptom set of the component is {no voltage output, no steady voltage, no varying voltage, large output ripple, oscillation clutter}. The fault cause set is {protection circuit fault, isolation circuit fault, control circuit fault, voltage-stabilizing circuit fault}. Using the fault diagnosis module obtained in this paper, 20 groups of training samples are taken for network training. Some samples are shown in Table 4.

Table 4
Part of training sample data

Serial number Fault symptom coding Fault cause coding

1 01010 0001

2 10000 1000

3 10011 0101

4 01110 0011

5 10100 1010

…… …… ……

Serial number	Fault symptom coding	Fault cause coding
1	01010	0001
2	10000	1000
3	10011	0101
4	01110	0011
5	10100	1010
……	……	……

The obtained training error is shown in Fig. 6.

Fig.6

Artificial intelligence fault diagnosis system BP neural network sample training error.

From Fig. 6, it can be seen that, after inputting 20 samples, the BP network iterates 17 steps to achieve a total error threshold of 0.05, which has a faster convergence rate. The training results of connection weight of the hidden layer and the input layer are shown in Table 5.

Table 5

BP network of artificial intelligence fault diagnosis system

Neuron	Weight					Threshold
	IN1	IN2	IN3	IN4	IN5
H0	–0.970000	0.840000	0.190000	0.370000	0.500000	0.716739
H1	–0.210000	–0.540000	–0.630000	–0.380000	0.180000	0.623318
H2	–0.160000	–0.900000	0.140000	–0.010000	–0.720000	–0.043745
H3	0.090000	–0.920000	0.0140000	–0.350000	0.650000	–0.882787
H4	–0.690000	0.270000	0.930000	–0.220000	0.190000	0.999627

From the analysis of Table 5, it can be obtained that, when the input is {no voltage output, large output ripple}, the diagnosis result is voltage-stabilizing circuit fault. Experimental data show that the BP network model realizes the fault pattern recognition and the learning function of fault knowledge. In addition, the more the training samples, the more comprehensive the fault pattern, the more reasonable the BP network model, the higher the accuracy of fault diagnosis.

The fault information of the related device can be extracted from the residual signal of the system when the artificial intelligent fault diagnosis of complex electronic equipment is carried out. Therefore, the main task of fault detection for complex electronic equipment is to obtain the residual signal from the system. For comparison between the designed system in this paper and traditional complex electronic equipment fault detection system based on image analysis theory, the time consuming of obtainment of residual signal in the process of artificial intelligent fault diagnosis is measured to verify the fault diagnosis rate of the designed system. Table 6 shows the time consuming of residual signal extraction of two systems.

Table 6

Two kinds of system residual signal extraction time (s)

Neuron	This paper system			Complex electronic equipment fault detection system based on traditional image analysis theory
	input layer	hidden layer	output layer	input layer	hidden layer	output layer
1	0.06	0.16	0.16	2.5	3.26	2.76
2	0.05	0.15	0.15	2.16	3.31	2.47
3	0.12	0.09	0.13	0.23	3.25	2.85
4	0.21	0.11	0.18	2.25	3.57	2.64
5	0.08	0.12	0.17	3.26	3.15	3.16
6	0.13	0.24	0.21	3.15	3.65	3.41
7	0.15	0.16	0.15	2.68	2.64	2.32
8	0.26	0.15	0.14	2.4	3.56	2.65
9	0.24	0.17	0.17	3.2	3.16	3.26
10	0.15	0.23	0.16	3.16	2.56	3.16
11	0.18	0.21	0.15	2.88	2.84	2.16
12	0.07	0.22	0.18	2.42	3.16	2.46
13	0.14	0.08	0.24	3.23	3.46	2.68
14	0.24	0.07	0.18	2.57	2.54	2.72
15	0.21	0.16	0.09	2.64	2.68	2.49
16	0.18	0.21	0.15	3.26	3.26	2.89
17	0.16	0.22	0.16	3.24	3.27	2.57
18	0.24	0.18	0.21	3.16	3.18	2.68
19	0.15	0.19	0.18	3.26	3.16	2.54
20	0.17	0.16	0.2	3.33	3.15	3.16

In Table 6, the time consuming of the residual signal obtained from the input layer, the hidden layer, and the output layer of 15 neurons in the complex electronic equipment is given. The time consuming of the proposed system is shortest. It can quickly make a different judgment on the residual signal in different neurons, so as to quickly and accurately respond to the fault cause of the electronic equipment [20 –22]. Comparison results show that the response time is shorter and the fault judgment rate is faster for the proposed system.

The neural network used in the experiment is shown in Fig. 2. The proposed artificial intelligent fault diagnosis is applied in the radar horn control relay panel of some complex electronic equipment. Through the simulation of the relay panel circuit, various fault phenomena and fault types are analyzed. 10 types of faults are taken as the output mode of neural network. The input is the voltage of 5 key nodes of a circuit board. Therefore, the number of input neurons in the designed neural network is 5, the number of output neurons is 10, and the number of units in the hidden layer is 16. The input and output samples of the neuron are shown in Table 7.

Table 7

Input and output sample

V1	V2	V3	V4	V5	fault type
11.264	14.134	10.631	14.779	11.331	Y1 (No fault)
0.171	14.78	0	14.78	0.248	Y2 (R1 open circuit)
11.234	10.465	10.452	10.881	10.882	Y3 (R2 open circuit)
11.264	14.134	10.833	14.779	11.536	Y4 (R3 open circuit)
11.264	14.78	14.757	14.78	13.355	Y5 (R4 open circuit)
0.741	14.653	0.153	14.78	0.002	Y6 (short circuit)
10.341	14.78	9.512	14.78	9.973	Y7 (BG1 Collector opening)
11.264	14.78	9.989	14.78	10.389	Y8 (BG1 Emitter opening)
11.333	11.341	12.305	13.02	13.012	Y9 (BG1 Base and collector short circuit)

In Table 7, Y1 represents no fault, and Y2 to Y10 represents various fault phenomena. If Yn (n = 2,…, 10) fault occurs, output Yn = 1 (n = 2,…, 10), and the rest is 0. If no fault occurs, all outputs are 0, andso on.

After normalizing the data in Table 7, the neural network model of the proposed system is used for model training. The number of the hidden layers is set to 16. E < 0.02. Learning rate λ and β = 0.05. Momentum factor η, δ = 0.75, and α = 0.05. After 20413 times of training, the relationship between the average training error and the step is shown in Fig. 7. With the data in Table 7, the outputs of BP neural network are shown in Table 8.

Fig.7

The relationship between the average training error and the step length.

Table 8

Output results of BP network

Y1	Y2	Y3	Y4	Y5	Y6	Y7	Y8	Y9	Y10
0.0001	0.0000	0.0006	0.0002	0.0015	0.0000	0.0000	0.0000	0.0004	0.0005
0.0000	0.9767	0.0000	0.0000	0.0068	0.0227	0.0000	0.0000	0.0000	0.0000
0.0003	0.0006	0.9995	0.0003	0.0004	0.0007	0.0018	0.0007	0.0079	0.0002
0.0782	0.0000	0.0000	0.9191	0.0018	0.0000	0.0000	0.0000	0.0002	0.0150
0.0007	0.0036	0.0000	0.0009	0.9851	0.0001	0.0002	0.0011	0.0000	0.0150
0.0063	0.0195	0.0033	0.0022	0.0000	0.9795	0.0021	0.0028	0.0007	0.0000
0.0000	0.0102	0.0008	0.0000	0.0000	0.0015	0.9731	0.0252	0.0000	0.0000
0.0208	0.0000	0.0000	0.0059	0.0005	0.0000	0.0244	0.9681	0.0000	0.0000
0.0019	0.0017	0.0096	0.0031	0.0004	0.0030	0.0001	0.0001	0.9898	0.0046
0.0008	0.0000	0.0000	0.0027	0.0161	0.0000	0.0000	0.0000	0.0083	0.9818

From Fig. 7, it can be known that, when the average error of training is required to be smaller, the longer the training time, the more the training steps and sometimes even not converging. When the requirement is met, the precision should be reduced to shorten the training time. According to the purpose of the project diagnosis and the function, E is set to 0.02 during the training.

The conversion rule of the output data of neural network to the fault: When the output maximum value Yn of a neural network is close to 1, and the other values are small, the system has no fault. From the analysis of Table 7, the correct rate of self-detection results of artificial intelligence fault diagnosis in this paper is 100%.

2 groups of actual diagnostic test data (shown in Table 9) is selected to verify the effectiveness of the proposed system.

Table 9

2 groups of experimental data

Group number	Sign
	V1	V2	V3	V4	V5
1	11.330	11.331	12.305	13.000	13.012
2	11.260	14.778	14.747	14.780	13.315

2 groups of experimental data are input into the network model, and the output results of the BP network are shown in Table 10.

Table 10

Output of the two groups of test data

Group number	Output results
	V1	V2	V3	V4	V5	V6	V7	V8	V9	V10
1	0.0001	0.0000	0.0084	0.0109	0.0000	0.0042	0.0000	0.0000	0.9883	0.0075
2	0.0003	0.0004	0.0002	0.0051	0.9867	0.0000	0.0000	0.0010	0.0000	0.0149

From the above two groups of output results, according to conversion rule, it can be seen that the Y9 output of the first groups is the largest, which is close to 1. Therefore, it can be judged that the Y9 fault occurs (BG1 base and collector short-circuit), which proves that the artificial intelligent fault diagnosis of the proposed system is correct. The Y5 output of the second group is largest, which is close to 1, while the output of the other units is smaller, which is ignored. Therefore, it can be judged that the Y5 fault occurs (R4 open-circuit). The actual verification is R4 open-circuits, so the diagnosis result is correct. Experimental results show that the precise fault diagnosis effect of the proposed system is better for complex electronic equipment.

4 Conclusions

The fault diagnosis and processing of complex electronic equipment involves both quantitative data and qualitative data. In fault diagnosis of certain equipment, the BP network diagnosis based on quantitative data has high accuracy and reliability, but it cannot be applied to the fault diagnosis based on qualitative data of the equipment. In this paper, the fault phenomenon with qualitative description is quantified by data preprocessing method. An intelligent fault diagnosis method based on BP neural network is proposed in this paper. It solves the problem of equipment fault diagnosis with qualitative data well.

Fault diagnosis of complex electronic equipment is an important research area of BP neural network. Based on the research results of BP neural network, the artificial intelligent fault diagnosis system for complex electronic is proposed in this paper, which can improve the accuracy of fault diagnosis and shorten the time consuming of fault diagnosis.

References

Zarković

and Stojković

, Analysis of artificial intelligence expert systems for power transformer condition monitoring and diagnostics, Electric Power Systems Research149 (2017), 125–136.

Mathan

and Yeung

, Extending the diagnostic capabilities of artificial intelligence-based instructional systems, Ai Magazine36(4) (2015), 51–60.

Adebayo

A.R.

, Abdulraheem

and Olatunji

S.O.

, Artificial intelligence based estimation of water saturation in complex reservoir systems, Journal of Porous Media18(9) (2015), 893–906.

Samarakou

, Papadakis

, Fylladitakis

E.D.

, et al., An open learning environment for the diagnosis, assistance and evaluation of students based on artificial intelligence, International Journal of Emerging Technologies in Learning9(3) (2014), 36–44.

Buyukyildiz

, Tezel

and Yilmaz

, Estimation of the change in lake water level by artificial intelligence methods, Water Resources Management28(13) (2014), 4747–4763.

Yin

and Zhao

, Fault diagnosis network design for vehicle on-board equipments of high-speed railway: A deep learning approach, Engineering Applications of Artificial Intelligence56 (2016), 250–259.

Bravo

C.E.

, Saputelli

, Rivas

, et al., State of the art of artificial intelligence and predictive analytics in the E&P industry: A technology survey, Spe Journal19(4) (2014), 547–563.

Guo

, Hu

and Peng

, Research of new strategies for improving CBR system, Artificial Intelligence Review42(1) (2014), 1–20.

Neagu

, Special issue on innovative techniques and applications of artificial intelligence, Expert Systems31(5) (2014), 409–410.

10.

Khaki

, Yusoff

and Islami

, Simulation of groundwater level through artificial intelligence system, Environmental Earth Sciences73(12) (2015), 8357–8367.

11.

Éric

and Mazure

, Introduction to the special issue on CSP technologies in artificial intelligence, Constraints20(4) (2015), 1–2.

12.

Khachatryan

G.A.

, Romashov

A.V.

, Khachatryan

A.R.

, et al., Reasoning mind genie 2: An intelligent tutoring system as a vehicle for international transfer of instructional methods in mathematics, International Journal of Artificial Intelligence in Education24(3) (2014), 333–382.

13.

Altman

, Artificial intelligence (AI) systems for interpreting complex medical datasets, Clinical Pharmacology & Therapeutics101(5) (2017), 585.

14.

Carrero

, Navas

, Malvárez

, et al., Artificial intelligence-based models to simulate land-use change around an estuary, Journal of Coastal Research70 (2014), 414–419.

15.

, Liu

, Li

, Furuta

and Peng

, Desulphurization characteristics of bamboo charcoal from sulfur solution, Saudi Journal of Biological Sciences24(1) (2017), 127–131.

16.

and Qiao

, Performance of Financial Expenditure in China’s basic science and math education: Panel Data Analysis Based on CCR Model and BBC Model, Eurasia Journal of Mathematics Science and Technology Education13(8) (2017), 5217–5224.

17.

Uysal

, Yilmaz

M.M.

and Ibikli

, On pointwise convergence of bivariate nonlinear singular integral operators, Kuwait Journal of Science44(2) (2017), 46–57.

18.

Zhao

Y.Z.

, Zhou

J.M.

and Xing

Q.R.

, A continuous hub location model with service constraints for shipping network design, Journal of Interdisciplinary Mathematics19(4) (2016), 681–695.

19.

Jiang

, Liu

and Lv

, Transportation system evaluation model based on DEA, Journal of Discrete Mathematical Sciences and Cryptography20(1) (2017), 115–124.

20.

Jin

, Liu

and Zhang

, Multiple solutions of fractional kirchhoff equations involving a critical nonlinearity, Discrete and Continuous Dynamical Systems-Series S11(3SI) (2018), 533–545.

21.

Gao

, Zali

M.R.

and Gao

, Ontology optimization tactics via distance calculating, Applied Mathematics and Nonlinear Sciences1 (2016), 154–169.

22.

Bazhenov

V.A.

, Pogorelova

O.S.

and Postnikova

T.G.

, Contact impact forces at discontinuous 2-DOF vibroimpact, Mathematics and Nonlinear Sciences1 (2016), 183–196.

Artificial intelligent fault diagnosis system of complex electronic equipment

Abstract

Keywords

1 Introduction

2 Artificial intelligent fault diagnosis system of complex electronic equipment based on BP neural network

2.1 System structure

2.2.1 BP neural network

Table 1 3-digit binary encoding table for the number of fault causes Serial number Number causes Binary coding 1 0 [000] 2 1 [001] 3 2 [010] 4 3 [011] 5 4 [100] 6 5 [101] 7 6 [110] 8 7 [111] 8 7 [111]

2.4.1 Design of database

3 Experimental analysis

Table 4 Part of training sample data Serial number Fault symptom coding Fault cause coding 1 01010 0001 2 10000 1000 3 10011 0101 4 01110 0011 5 10100 1010 …… …… ……

References

Table 1
3-digit binary encoding table for the number of fault causes

Serial number Number causes Binary coding

1 0 [000]

2 1 [001]

3 2 [010]

4 3 [011]

5 4 [100]

6 5 [101]

7 6 [110]

8 7 [111]

8 7 [111]

Table 4
Part of training sample data

Serial number Fault symptom coding Fault cause coding

1 01010 0001

2 10000 1000

3 10011 0101

4 01110 0011

5 10100 1010

…… …… ……