Multi-source information fusion coal cutter cutting unit fault prediction system construction method

Abstract

The invention provides a multi-source information fusion coal cutter cutting unit fault prediction system construction method, relates to the technical field of coal cutter cutting unit fault prediction, and according to the fault classification problem of cutting unit key part failure, a cutting unit rigid body model is established, and a cutting unit virtual prototype model is established by using Adams software. And respectively importing the fault-free model and the fault-containing model into Adams, completing real mapping of a physical model machine, and respectively testing, extracting and analyzing axial and radial stress signals of test shafts of the fault-free model and the fault-containing model in the working process of the cutting part of the coal mining machine. Based on theoretical methods of wavelet packet decomposition and a linear travel method, experimental data of different working conditions and different fault forms are processed, energy of sub-bands after wavelet packet decomposition is calculated, and a coal mining machine cutting unit fault prediction data processing module is built; the Elman neural network is used for carrying out fault mode prediction on a coal cutter test shaft signal, and a coal cutter cutting unit fault prediction data identification module is built.

Description

Technical Field

[0001] The present invention relates to the technical field of fault prediction of a cutting part of a coal mining machine, and in particular to a method for constructing a fault prediction system of a cutting part of a coal mining machine by fusion of multi-source information. Background Art

[0002] 90% of my country's coal is mined underground. The cutting and crushing of complex interlayer coal and rock in underground is an inevitable phenomenon of coal mining and excavation, and is the result of the coupling of multiple factors. The interaction between the coal mining machine and the cut coal and rock, the gradual characteristics of the power transmission system, the kinematic parameters in the working process, and the complex and diverse occurrence conditions of the cut coal and rock will have a direct or indirect impact on its cutting and crushing process, making the coal mining machine fault prediction technology poor in practicality and universality, at the cost of a large amount of energy consumption, low efficiency and poor equipment dynamic reliability. At present, the state has successively issued a series of relevant policies such as the "Notice of the National Energy Administration on Further Accelerating the Construction of Intelligent Coal Mines and Promoting High-quality Development of Coal", which has put forward new challenges to the design of intelligent coal mining working faces. Designing intelligent, efficient and reliable coal mining machines is one of the important technologies for the comprehensive and accurate perception of the coal resource mining environment and the coordinated coupling of equipment operation. Improving the reliability of the coal mining machine power transmission system and the deep integration of the new generation of information technology is the key technical bottleneck to master the new round of scientific and technological revolution and change the way of mineral resource development and utilization.

[0003] Affected by complex loads and a large number of impacts, the occurrence of gear and bearing failures in the cutting part of the coal mining machine can easily cause the coal mining machine to shut down, seriously affecting the safe production of coal mines. Timely and accurate fault diagnosis is an effective way to improve productivity and reduce the accident rate. In previous studies, the experimental fault judgment was unclear, the prediction data source was single, and the location of the fault could not be accurately determined, making it difficult to achieve accurate prediction and decision-making of the fault in the cutting part of the coal mining machine. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention provides a method for constructing a coal mining machine cutting part fault prediction system with multi-source information fusion; it aims to utilize the fusion of multi-source data information such as discrete element system, finite element system, data processing and perception system, to establish a coal mining machine cutting part fault diagnosis system integrating model construction, data acquisition, data processing and data perception, so as to realize autonomous prediction and judgment of coal mining machine cutting part faults.

[0005] A method for constructing a multi-source information fusion coal mining machine cutting part fault prediction system comprises the following steps:

[0006] Step 1: Select the missing rolling element, inner ring crack and outer ring crack of the rocker arm eight-axis bearing as fault samples, build a fault model, and conduct a test experiment;

[0007] Step 1.1: Establish rigid body model and add constraints;

[0008] Through the virtual prototype analysis software Adams and Pro / E interface software, the three-dimensional solid model of the cut part with the faulty parts is imported into Adams to form a rigid body model. The rigid body model in Adams is added with rotational pairs, translation pairs, contact and friction as constraints according to the working principle of the equipment, and the contact stiffness is calculated. The stiffness of key components is added according to the calculation results.

[0009] Step 1.2: Establishment of rigid-flexible coupling model;

[0010] The test shaft and the faulty parts in the three-dimensional solid model of the cutting part with the faulty parts are made into flexible bodies to establish a fault model; the axial and radial force signals of the test shaft are used as diagnostic parameters to perform flexible processing on the eight-axis bearings of the rocker arm and the four axes of the rocker arm respectively;

[0011] Step 2: Acquisition of fault prediction data signal;

[0012] Based on the data coupling method of discrete element and finite element, a dynamic bidirectional coupling model of the cutting part of the coal mining machine is built during its operation.

[0013] According to the design requirements, the drum speed, rotation cycle, simulation time and step length are set, and the three-dimensional solid models of the cutting parts with and without faults are imported into Adams for grouping and virtual prototype simulation. The grouping is shown in Table 1. After the simulation, the post-processing data analysis module of Adams is entered to extract the axial force and radial force of each test axis. Each simulation corresponds to 4 test axes, with a total of eight data.

[0014] Table 1 Experimental groups

[0015]

[0016] Step 3: Processing of fault prediction data signals;

[0017] The relationship between the signals of each test axis and the working conditions of the coal mining machine is analyzed, and the axial and radial forces of each test axis are extracted and analyzed in sections by using the wavelet packet decomposition method, and the energy value of the wavelet packet subband is calculated by using the linear stroke method.

[0018] Step 3.1: Feature processing of data signals;

[0019] The CoifN wavelet system is selected as the wavelet basis function for wavelet packet decomposition, N = 1, 2, 3, 4, 5, N is the sequence number of the wavelet function series, and the axial force and radial force of each test shaft are decomposed by wavelet packets respectively, and 16 sub-band data are obtained for each test shaft data;

[0020] Step 3.2: Calculation of data signal capability;

[0021] The line-travel method is used to quantize each sub-band signal after wavelet packet decomposition, and the sum of the distance between two adjacent points is used as the energy value of the signal.

[0022] E 2 =X1 2 +(X2-X1) 2 +(X3-X2) 2 +…+(X n -X n-1 ) 2

[0023] Where E is the energy value of each sub-band signal, X n It is the detail signal of the nth wavelet packet decomposition;

[0024] The signal energy of each sub-band signal after wavelet packet decomposition of all working conditions and fault modes is calculated according to the linear travel method. Table 2 shows the sub-band energy of wavelet packet decomposition corresponding to the fault-free test shaft signals under full coal working conditions. The same processing is performed on the signals of other various fault forms to obtain the wavelet packet sub-band energy value of each group of experimental test shaft stress data.

[0025] Table 2

[0026]

[0027] Step 4: Perception of fault prediction data signals;

[0028] Establish the perception Elman neural network of the fault prediction data signal of the cutting part of the coal mining machine;

[0029] The perception Elman neural network of the coal mining machine cutting part fault prediction data signal is a dynamic regression neural network, and the mathematical model of the perception Elman neural network of the coal mining machine cutting part fault prediction data signal is as follows:

[0030]

[0031] Where x(u) is the input value of the neural network; y(u) is the output value of the neural network; x y (u) is the hidden layer output; x c (u) is the output of the receiving layer, u is the number of nodes in each layer; f(u) is the hidden layer transfer function; g(u) is the output layer transfer function; α is the receiving layer node unit vector, b1 and b2 are the thresholds of the hidden layer and the output layer respectively; W i is the weight of the i-th layer. The perception Elman neural network of the fault prediction data signal of the coal mining machine cutting part minimizes the evaluation value through the gradient search theory. The evaluation value is the mean square error between the actual output and the target output:

[0032]

[0033] Where C is the mean square error; n is the number of iterations, which is the number of samples minus 1; y(t) and are the actual output and target output of the network respectively;

[0034] The input vector x1(u) of the perceptual Elman neural network of the coal mining machine cutting part fault prediction data signal is:

[0035]

[0036] x1(u)={T 1x (u), T 1y (u), T 2x (u), T 2y (u), T 3x (u), T 3y (u), T 4x (u), T 4y (u)} T (3)

[0037] Where T ix Indicates the axial signal of the test shaft, T iy Indicates the radial signal of the test axis, i is the test axis number, i = 1, 2, 3, 4, f ix (t) is the single-cycle experimental axial data of test axis i, f iy (t) is the radial data of a single-cycle experiment on the test axis i; E (a,b) (x) represents the subband energy value of the wavelet packet decomposition, a is the decomposition layer number, and b is the node number; the final x1(u) is a 128*1 vector.

[0038] The output states of the perceptual Elman neural network of the coal mining machine cutting part fault prediction data signal include two conditions: full coal condition and gangue condition. The corresponding network output is shown in Table 3.

[0039] Table 3

[0040]

[0041] With formula (3) as the neural network input and Table 3 as the neural network output, the working condition classification neural network model NET1 is established. After network model accuracy training, the working condition classification neural network with the highest accuracy is obtained, that is, the three-layer neural network of 38-9-2;

[0042] The neural network output vectors corresponding to various fault types are set according to the fault type, as shown in Table 4:

[0043] Table 4

[0044]

[0045]

[0046] Corresponding to the four fault forms under full coal working conditions, the parameters are optimized, and after the network model accuracy training, the coal mining machine cutting part fault prediction data signal perception 20-13-11Elman neural network NET2 is established;

[0047] Corresponding to the four fault forms under the condition of gangue, the parameters are optimized, and after the network model accuracy training, the 30-50-22-11 Elman neural network NET3 for predicting the fault data signal perception of the cutting part of the coal mining machine is established;

[0048] A virtual prototype experiment was conducted again for various fault forms under full coal working condition and gangue intercalation working condition, and the test data obtained by the simulation was input into the neural network NET1 for working condition classification. The output results of the neural network are shown in Table 5. According to the output vector, the test vectors Test1-Test4 are determined to be full coal working condition, and Test5-Test8 are determined to be gangue intercalation working condition;

[0049] Table 5

[0050]

[0051] According to the diagnosis results of neural network NET1 shown in Table 5, the test data are input into neural networks NET2 and NET3 for fault form classification. The output and classification results of neural networks NET2 and NET3 are shown in Tables 6 and 7 respectively.

[0052] Table 6

[0053]

[0054] Table 7

[0055]

[0056] Step 5: Integrated packaging of multi-source data information

[0057] Relying on the Matlab / Simulink simulation control platform, the processing of the fault prediction data signal in step 3 and the perception of the fault prediction data signal in step 4 are integrated and packaged, and multi-field coupling is formed with the acquisition of the fault prediction data signal in step 2 to build a virtual coal mining machine cutting part fault prediction feedback system with multi-source data fusion to achieve real-time data sharing, online perception and diagnosis.

[0058] The beneficial effects of adopting the above technical solution are:

[0059] The present invention provides a method for constructing a coal mining machine cutting part fault prediction system with multi-source information fusion. In previous studies, the experimental fault judgment was unclear, the prediction data source was single, and the fault location could not be accurately judged, making it difficult to achieve accurate prediction and decision-making of the coal mining machine cutting part fault. The present invention integrates multi-source data information such as discrete element system, finite element system, data processing and perception system, and establishes a coal mining machine cutting part fault diagnosis system that integrates model construction, data acquisition, data processing and data perception, so as to improve the accurate and autonomous prediction and judgment capabilities of coal mining machine cutting part faults. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 It is a structural schematic diagram of the rigid body model of the present invention;

[0061] Figure 2 A schematic diagram of the constraint application of the rocker arm transmission system of the present invention;

[0062] Figure 3 It is a schematic diagram of the roller constraint application of the present invention;

[0063] Figure 4 This is a schematic diagram of the flexible treatment of the eight-axis bearings and four axes of the rocker arm of the present invention;

[0064] Wherein (a) is a schematic diagram of the test axis position and the flexible processing of the test axis of the virtual prototype of the present invention, (b) is a schematic diagram of the outer ring defect flexible body, and (c) is a schematic diagram of the rolling element defect flexible body;

[0065] Figure 5 It is a schematic diagram of a dynamic bidirectional coupling model of the cutting part of the coal mining machine of the present invention during operation;

[0066] Figure 6 It is a schematic diagram of the force information of the shaft 4 in the radial direction when the coal mining machine of the present invention is in a fault-free state of cutting full coal;

[0067] Where (a) is the original signal, (b) is the schematic diagram of subband 1 after wavelet packet decomposition, and (c) is the schematic diagram of subband 2 after wavelet packet decomposition;

[0068] Figure 7 A schematic diagram of the neural network classification process for signals of the present invention;

[0069] Figure 8 It is a schematic diagram of the structure of the perceptual Elman neural network of the fault prediction data signal of the cutting part of the coal mining machine of the present invention;

[0070] Fig. 9 This is a structural schematic diagram of the working condition classification neural network model NET1 of the present invention;

[0071] Fig.10This is a schematic diagram of the structure of the Elman neural network NET2 of the fault prediction data signal perception 20-13-11 of the coal mining machine cutting part of the present invention;

[0072] Fig.11 This is a schematic diagram of the structure of the NET3 Elman neural network for predicting the fault of the cutting part of the coal mining machine according to the present invention.

[0073] Fig.12 It is a structural schematic diagram of the multi-source data fusion virtual coal mining machine cutting part fault prediction feedback system of the present invention. DETAILED DESCRIPTION

[0074] The specific implementation of the present invention is further described in detail below in conjunction with the accompanying drawings and examples. The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention.

[0075] A method for constructing a multi-source information fusion coal mining machine cutting part fault prediction system comprises the following steps:

[0076] Step 1: Multiple failures of coal mining machines and maintenance records of coal mining machines show that the common failed parts of the cutting part of coal mining machines are low-speed and heavy-load rolling bearings at the output end. Therefore, this method selects the missing rolling element, inner ring cracks and outer ring cracks of the rocker arm eight-axis bearing (large load) as fault samples, builds a fault model, and conducts test experiments;

[0077] Step 1.1: Establish rigid body model and add constraints;

[0078] Through the virtual prototype analysis software Adams and Pro / E interface software MECH / pro2005, the three-dimensional solid model of the cut part with the faulty parts is imported into Adams to form a rigid body model, such as Figure 1 As shown. According to the working principle of the equipment, add rotational pairs, translational pairs, contact and friction as constraints to the rigid body model in Adams, as shown Figure 2 , Figure 3 As shown, the contact stiffness is calculated and the stiffness of key components is added based on the calculation results.

[0079] Step 1.2: Establishment of rigid-flexible coupling model;

[0080] The test shaft and the faulty parts in the three-dimensional solid model of the cutting part with the faulty parts are made into flexible bodies to establish the fault model; the axial and radial force signals of the test shaft are used as diagnostic parameters, and the eight-axis bearings of the rocker arm and the four axes of the rocker arm are processed flexibly, such as Figure 4 In this embodiment Figure 4 (a) is a schematic diagram of the test axis position and test axis flexibility processing of the virtual prototype of the present invention, Figure 4 (b) is a schematic diagram of the outer ring defect flexible body. Figure 4 (c) is a schematic diagram of a rolling element defective flexible body;

[0081] Step 2: Acquisition of fault prediction data signal;

[0082] Based on the data coupling method of discrete element and finite element, a dynamic bidirectional coupling model of the cutting part of the coal mining machine is built during its working process; Figure 5 shown.

[0083] According to the design requirements, the drum speed is 87.2r / min, the rotation cycle is 0.7187s, the simulation time is set to 0.95s, and the step size is 0.0005s (corresponding to the sampling frequency of 2000Hz). The three-dimensional solid models of the cutting parts with and without faults are imported into Adams and grouped for virtual prototype simulation. The grouping is shown in Table 1. After the simulation, the post-processing data analysis module of Adams is entered to extract the axial force and radial force of each test axis, and export them as *.tab data files that can be read by Matlab. Each simulation corresponds to 4 test axes, with a total of eight data. To ensure the accuracy of the diagnostic system, each group of virtual prototype simulations is repeated 20 times.

[0084] Table 1 Experimental groups

[0085]

[0086] Step 3: Processing of fault prediction data signals;

[0087] The relationship between the signals of each test axis and the working conditions of the coal mining machine is analyzed, and the axial and radial forces of each test axis are extracted and analyzed in sections by using the wavelet packet decomposition method, and the energy value of the wavelet packet subband is calculated by using the linear stroke method.

[0088] Step 3.1: Feature processing of data signals;

[0089] The CoifN wavelet system is selected as the wavelet basis function for wavelet packet decomposition, N = 1, 2, 3, 4, 5, the wavelet function has a series of CoifN, N is the sequence number of the wavelet function series, and the axial force and radial force of each test shaft are respectively decomposed by wavelet packets, and 16 sub-band data are obtained for each test shaft data; Figure 6 The figure shows the force information of the shaft 4 in the radial direction when the coal mining machine is in the full coal cutting state without any fault. Figure 6 (a) is the original signal, Figure 6 (b) is a schematic diagram of subband 1 decomposed by wavelet packet. Figure 6 (c) Schematic diagram of subband 2 of wavelet packet decomposition;

[0090] Step 3.2: Calculation of data signal capability;

[0091] The line-travel method is used to quantize each sub-band signal after wavelet packet decomposition, and the sum of the distance between two adjacent points is used as the energy value of the signal.

[0092] E 2 =X1 2 +(X2-X1) 2 +(X3-X2) 2 +…+(X n -X n-1 ) 2

[0093] Where E is the energy value of each sub-band signal, X n It is the detail signal of the nth wavelet packet decomposition;

[0094] The signal energy of each sub-band signal after wavelet packet decomposition of all working conditions and fault modes is calculated according to the linear travel method. Table 2 shows the sub-band energy of wavelet packet decomposition corresponding to the fault-free test shaft signals under full coal working conditions. The same processing is performed on the signals of other various fault forms to obtain the wavelet packet sub-band energy value of each group of experimental test shaft stress data.

[0095] Table 2

[0096]

[0097] Step 4: Perception of fault prediction data signals;

[0098] Establish the perception Elman neural network of the fault prediction data signal of the cutting part of the coal mining machine; Figure 7 The figure shows the process of establishing a neural network classification for the signal.

[0099] The perception Elman neural network of the coal mining machine cutting part fault prediction data signal is a dynamic regression neural network. Figure 8 The network structure diagram is shown. The mathematical model of the perception Elman neural network of the fault prediction data signal of the coal mining machine cutting part is as follows:

[0100]

[0101] Where x(u) is the input value of the neural network; y(u) is the output value of the neural network; x y (u) is the hidden layer output; x c (u) is the output of the receiving layer, u is the number of nodes in each layer; f(u) is the hidden layer transfer function; g(u) is the output layer transfer function; α is the receiving layer node unit vector, b1 and b2 are the thresholds of the hidden layer and the output layer respectively; W i is the weight of the i-th layer. The perception Elman neural network of the fault prediction data signal of the coal mining machine cutting part minimizes the evaluation value through the gradient search theory. The evaluation value is the mean square error between the actual output and the target output:

[0102]

[0103] Where C is the mean square error; n is the number of iterations, which is the number of samples minus 1; y(t) and are the actual output and target output of the network respectively;

[0104] The input vector x1(u) of the perceptual Elman neural network of the coal mining machine cutting part fault prediction data signal is:

[0105]

[0106] x1(u)={T 1x (u), T 1y (u), T 2x (u), T 2y (u), T 3x (u), T 3y (u), T 4x (u), T 4y (u)} T (3)

[0107] Where T ix Indicates the axial signal of the test shaft, T iy Indicates the radial signal of the test axis, i is the test axis number, i = 1, 2, 3, 4, f ix (t) is the single-cycle experimental axial data of test axis i, f iy (t) is the radial data of a single-cycle experiment on the test axis i; E (a,b) (x) represents the subband energy value of the wavelet packet decomposition, a is the decomposition layer number, and b is the node number; the final x1(u) is a 128*1 vector.

[0108] The output states of the perceptual Elman neural network of the coal mining machine cutting part fault prediction data signal include two conditions: full coal condition and gangue condition. The corresponding network output is shown in Table 3.

[0109] Table 3

[0110]

[0111] With formula (3) as the neural network input and table 3 as the neural network output, the working condition classification neural network model NET1 is established. After network model accuracy training, the working condition classification neural network with the highest accuracy is obtained, that is, the three-layer neural network of 38-9-2, where 38-9-2 is the number of unit nodes of the middle hidden coal seam, as shown in Fig. 9 shown.

[0112] The neural network output vectors corresponding to various fault types are set according to the fault type, as shown in Table 4:

[0113] Table 4

[0114]

[0115] Corresponding to the four fault forms under full coal working conditions, parameter optimization is carried out, and after network model accuracy training, the coal mining machine cutting part fault prediction data signal perception 20-13-11Elman neural network NET2 is established. Fig.10 shown.

[0116] Corresponding to the four fault forms under the gangue-binding condition, the parameters are optimized, and the network model accuracy training is carried out to establish the 30-50-22-11Elman neural network NET3 for fault prediction data signal perception of the cutting part of the coal mining machine. Fig.11 shown.

[0117] A virtual prototype experiment was conducted again for various fault forms under full coal working condition and gangue intercalation working condition, and the test data obtained by the simulation was input into the neural network NET1 for working condition classification. The output results of the neural network are shown in Table 5. According to the output vector, the test vectors Test1-Test4 are determined to be full coal working condition, and Test5-Test8 are determined to be gangue intercalation working condition;

[0118] Table 5

[0119]

[0120] According to the diagnosis results of neural network NET1 shown in Table 5, the test data are input into neural networks NET2 and NET3 for fault form classification. The output and classification results of neural networks NET2 and NET3 are shown in Tables 6 and 7 respectively.

[0121] Table 6

[0122]

[0123] Table 7

[0124]

[0125]

[0126] Step 5: Integrated packaging of multi-source data information

[0127] Relying on the Matlab / Simulink simulation control platform, the processing of the fault prediction data signal in step 3 and the perception of the fault prediction data signal in step 4 are integrated and packaged, and multi-domain coupling is formed with the acquisition of the fault prediction data signal in step 2 to build a virtual coal mining machine cutting unit fault prediction feedback system with multi-source data fusion to realize real-time data sharing, online perception and diagnosis. Fig.12 shown.

[0128] The above description is only a preferred embodiment of the present disclosure and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present disclosure is not limited to the technical solutions formed by a specific combination of the above-mentioned technical features, but should also cover other technical solutions formed by any combination of the above-mentioned technical features or their equivalent features without departing from the above-mentioned inventive concept. For example, the above-mentioned features are replaced with the technical features with similar functions disclosed in the embodiments of the present disclosure (but not limited to) to form a technical solution.

Claims

1. A method for constructing a fault prediction system for the cutting part of a coal mining machine based on multi-source information fusion, characterized in that: The following steps are involved: Step 1: Select the missing rolling element, inner ring crack and outer ring crack of the rocker arm eight-axis bearing as fault samples, build a fault model, and conduct a test experiment; Step 2: Acquisition of fault prediction data signal; Based on the data coupling method of discrete element and finite element, a dynamic bidirectional coupling model of the cutting part of the coal mining machine is built during its operation. Step 3: Processing of fault prediction data signals; The relationship between the signals of each test axis and the working conditions of the coal mining machine is analyzed, and the axial and radial forces of each test axis are extracted and analyzed in sections by using the wavelet packet decomposition method, and the energy value of the wavelet packet subband is calculated by using the linear stroke method. Step 4: Perception of fault prediction data signals; Establishing the perception Elman neural network of fault prediction data signals of the cutting part of the coal mining machine; Step 5: Integrated packaging of multi-source data information; Relying on the Matlab / Simulink simulation control platform, the processing of the fault prediction data signal in step 3 and the perception of the fault prediction data signal in step 4 are integrated and packaged, and multi-field coupling is formed with the acquisition of the fault prediction data signal in step 2 to build a virtual coal mining machine cutting part fault prediction feedback system with multi-source data fusion to achieve real-time data sharing, online perception and diagnosis.

2. The method for constructing a multi-source information fusion coal mining machine cutting part fault prediction system according to claim 1, characterized in that: The step 1 specifically comprises the following steps: Step 1.1: Establish rigid body model and add constraints; The 3D solid model of the cut part with the faulty parts is imported into Adams through the virtual prototype analysis software Adams and Pro / E interface software to form a rigid body model. According to the working principle of the equipment, the rigid body model in Adams is added with rotational pairs, translation pairs, contact and friction as constraints, and the contact stiffness is calculated. The stiffness of key components is added according to the calculation results. Step 1.2: Establishment of rigid-flexible coupling model; The test shaft and faulty parts in the three-dimensional solid model of the cutting part with faulty parts are made into flexible bodies to establish a fault model. The axial and radial force signals of the test shaft are used as diagnostic parameters, and the eight-axis bearings of the rocker arm and the four axes of the rocker arm are made flexible.

3. The method for constructing a multi-source information fusion coal mining machine cutting part fault prediction system according to claim 1, characterized in that: The step 2 specifically includes: The drum speed, rotation cycle, simulation time and step length are set according to the design requirements, and the three-dimensional solid models of the cutting parts with and without faults are imported into Adams and grouped for virtual prototype simulation. The grouping is shown in Table 1: Table 1 After the simulation is completed, enter the post-processing data analysis module of Adams to extract the axial force and radial force of each test axis. Each simulation corresponds to 4 test axes, with a total of eight data.

4. The method for constructing a multi-source information fusion coal mining machine cutting part fault prediction system according to claim 1, characterized in that: The step 3 specifically includes the following steps: Step 3.1: Feature processing of data signals; The CoifN wavelet system is selected as the wavelet basis function for wavelet packet decomposition, N = 1, 2, 3, 4, 5, N is the sequence number of the wavelet function series, and the axial force and radial force of each test shaft are decomposed by wavelet packets respectively, and 16 sub-band data are obtained for each test shaft data; Step 3.2: Calculation of data signal capability; The line-travel method is used to quantize each sub-band signal after wavelet packet decomposition, and the sum of the distance between two adjacent points is used as the energy value of the signal. E 2 =X1 2 +(X2-X1) 2 +(X3-X2) 2 +…+(X n -X n-1 ) 2 Where E is the energy value of each sub-band signal, X n It is the detail signal of the nth wavelet packet decomposition; According to the line travel method, the signal energy of each sub-band signal after wavelet packet decomposition of all working conditions and fault modes is calculated. Table 2 shows the wavelet packet decomposition sub-band energy corresponding to each test axis signal in the form of no fault under full coal working conditions: Table 2 The same processing is performed on the signals of other fault forms to obtain the wavelet packet sub-band energy value of each group of experimental test shaft force data.

5. The method for constructing a multi-source information fusion coal mining machine cutting part fault prediction system according to claim 1, characterized in that: The perception Elman neural network of the coal mining machine cutting part fault prediction data signal in step 4 is a dynamic regression neural network. The mathematical model of the perception Elman neural network of the coal mining machine cutting part fault prediction data signal is as follows: Where x(u) is the input value of the neural network; y(u) is the output value of the neural network; x y (u) is the hidden layer output; x c (u) is the output of the receiving layer, u is the number of nodes in each layer; f(u) is the hidden layer transfer function; g(u) is the output layer transfer function; α is the receiving layer node unit vector, b1 and b2 are the thresholds of the hidden layer and the output layer respectively; W i is the weight of the i-th layer. The perception Elman neural network of the fault prediction data signal of the coal mining machine cutting part minimizes the evaluation value through the gradient search theory. The evaluation value is the mean square error between the actual output and the target output: Where C is the mean square error; n is the number of iterations, which is the number of samples minus 1; y(t) and are the actual output and target output of the network respectively.

6. The method for constructing a multi-source information fusion coal mining machine cutting part fault prediction system according to claim 5, characterized in that: The input vector x1(u) of the perceptual Elman neural network of the coal mining machine cutting part fault prediction data signal is: x1(u)={T 1x (u),T 1y (u),T 2x (u),T 2y (u),T 3x (u),T 3y (u),T 4x (u),T 4y (u)} T (3) Where T ix Indicates the axial signal of the test shaft, T iy Indicates the radial signal of the test axis, i is the test axis number, i = 1, 2, 3, 4, f ix (t) is the single-cycle experimental axial data of test axis i, f iy (t) is the radial data of a single-cycle experiment on the test axis i; E (a,b) (x) represents the subband energy value of the wavelet packet decomposition, a is the decomposition layer number, and b is the node number; the final x1(u) is a 128*1 vector; The output states of the perception Elman neural network of the fault prediction data signal of the coal mining machine cutting part include two conditions: full coal working condition and gangue intercalation working condition. The corresponding network output is shown in Table 3. Table 3 With formula (3) as the neural network input and Table 3 as the neural network output, the working condition classification neural network model NET1 is established. After network model accuracy training, the working condition classification neural network with the highest accuracy is obtained, that is, the three-layer neural network of 38-9-2; The neural network output vectors corresponding to various fault types are set according to the fault type, as shown in Table 4: Table 4 Corresponding to the four fault forms under full coal working conditions, the parameters are optimized, and after the network model accuracy training, the coal mining machine cutting part fault prediction data signal perception 20-13-11Elman neural network NET2 is established; Corresponding to the four fault forms under the condition of gangue, the parameters are optimized, and after the network model accuracy training, the 30-50-22-11 Elman neural network NET3 for predicting the fault data signal perception of the cutting part of the coal mining machine is established; A virtual prototype experiment was conducted again for various fault forms under full coal working condition and gangue intercalation working condition, and the test data obtained by the simulation was input into the neural network NET1 for working condition classification. The output results of the neural network are shown in Table 5. According to the output vector, the test vectors Test1-Test4 are determined to be full coal working condition, and Test5-Test8 are determined to be gangue intercalation working condition. Table 5 According to the diagnosis results of neural network NET1 shown in Table 5, the test data are input into neural networks NET2 and NET3 for fault form classification, as shown in Tables 6 and 7: Table 6 Table 7 Tables 6 and 7 are the output and classification results of neural networks NET2 and NET3 respectively.

Images