Fault diagnosis method and system for off-road forklift transmission system based on artificial intelligence

Abstract

The present invention discloses a fault diagnosis method and system for an off-road forklift transmission system based on artificial intelligence, the method comprising: obtaining real-time operating state data of an off-road forklift transmission system to obtain first operating state data; performing multi-scale feature extraction and fusion on the first operating state data to obtain a first comprehensive feature matrix, and then obtaining a second comprehensive feature matrix through center of gravity compensation; identifying the key time steps of the second comprehensive feature matrix to obtain a third comprehensive feature matrix, performing forward feature and backward feature extraction and fusion on the third comprehensive feature matrix to obtain a bidirectional time feature vector; obtaining each operating state integral based on the real-time operating state data, the bidirectional time feature vector and the second comprehensive feature matrix; obtaining the fault integral of each fault type based on each operating state integral, and then obtaining the fault diagnosis result. The method of the present application can improve the accuracy and reliability of fault diagnosis of an off-road forklift transmission system.

Description

Technical Field

[0001] The present invention relates to the field of off-road forklifts, and in particular to an artificial intelligence-based off-road forklift transmission system fault diagnosis method and system. Background Art

[0002] With the development of industry and agriculture, off-road forklifts will be used preferentially in complex working conditions and the frequency of use will gradually increase. Therefore, the frequency of transmission system failure will also increase with the increase in frequency of use. As the core part of off-road forklift power transmission, the transmission system includes multiple subsystems or components such as clutch, gearbox, drive shaft, differential, universal joint, half shaft, coupling, gear, hydraulic system and bearing. These components work together to effectively transmit the power of the engine or motor to the drive wheel or actuator to achieve the normal operation of the forklift.

[0003] However, as the use environment of off-road forklifts becomes increasingly complex, these components will face higher loads and harsher working conditions, and are therefore prone to various faults, such as wear, fatigue, and failure. For example, under high loads and frequent use, components such as drive shafts, clutches, and differentials may generate fault signals such as abnormal vibration, temperature rise, and pressure changes. If these fault signals are not diagnosed in a timely and accurate manner, the overall performance of the transmission system will decline, and even cause serious equipment failures of the off-road forklift.

[0004] Traditional fault diagnosis methods usually rely on single sensor signals or simple statistical analysis, which makes it difficult to accurately capture the relationship between various signals in complex systems and is prone to misjudgment or missed judgment. In addition, there are limitations in responding to the real-time fault diagnosis needs of off-road forklifts under variable working conditions, resulting in high maintenance costs and adverse effects on the long-term service life of off-road forklifts. Therefore, there is an urgent need for a method that can efficiently and accurately diagnose transmission system faults to improve the reliability and accuracy of off-road forklift transmission system fault diagnosis. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention provides an artificial intelligence-based fault diagnosis method for an off-road forklift transmission system.

[0006] In order to solve the above technical problems, the present invention is solved by the following technical solutions:

[0007] A fault diagnosis method for an off-road forklift transmission system based on artificial intelligence comprises the following steps:

[0008] Acquire real-time operating status data of the transmission system of the off-road forklift, pre-process the real-time operating status data, and obtain first operating status data, wherein the first operating status data at least includes first center of gravity data, first vibration data, and first strain data;

[0009] Perform multi-scale feature extraction on the first operating state data to obtain a multi-scale feature matrix and fuse the multi-scale feature matrix to obtain a first comprehensive feature matrix;

[0010] Performing gravity compensation on the first vibration data and the first strain data by using the first gravity center data to update the first comprehensive feature matrix to obtain a second comprehensive feature matrix;

[0011] Based on the global attention mechanism, the key time steps of the second comprehensive feature matrix are identified, and then the second comprehensive feature matrix is ​​weighted to obtain a third comprehensive feature matrix, and forward features and backward features are extracted and fused on the third comprehensive feature matrix to obtain a bidirectional time feature vector;

[0012] A weight coefficient of each operating state data is obtained based on the change amplitude of the real-time operating state data, and each operating state integral is obtained by combining the bidirectional time feature vector and the second comprehensive feature matrix;

[0013] Each operating state integral is judged and adjusted, and the weight of each operating state integral in each fault type is determined, thereby obtaining the fault integral of each fault type, and obtaining the fault diagnosis result according to the fault integral of each fault type.

[0014] As an implementable method, performing multi-scale feature extraction on the first operating state data to obtain a multi-scale feature matrix and fusing the matrix to obtain a first comprehensive feature matrix includes the following steps:

[0015] Extracting multi-scale features of each running state data with different scale parameters and different time steps from each running state data in the first running state data by wavelet transform based on different scale parameters and different time steps;

[0016] Based on the multi-scale features, a first multi-scale feature matrix corresponding to each operating state is obtained, and the first multi-scale feature matrix corresponding to each operating state is fused to obtain a first comprehensive feature matrix;

[0017] Among them, wavelet transform is expressed as follows:

[0018]

[0019] in, represents the wavelet transform coefficients, represents the scale parameter, represents the time shift parameter, represents the first running status data, represents the mother wavelet function, Represents the time step.

[0020] As an implementable method, performing gravity compensation on the first vibration data and the first strain data by using the first gravity center data to update the first comprehensive feature matrix to obtain the second comprehensive feature matrix includes the following steps:

[0021] Performing low-pass filtering on the first centroid data to obtain second centroid data;

[0022] Constructing a gravity compensation model, performing gravity compensation on the first vibration data and the first strain data through the second gravity data, and obtaining the second vibration data and the second strain data;

[0023] Performing multi-scale feature extraction on the second vibration data and the second strain data to obtain a second multi-scale vibration feature matrix and a second multi-scale strain feature matrix;

[0024] The first comprehensive feature matrix is ​​updated by using the second multi-scale vibration feature matrix and the second multi-scale strain feature matrix to obtain a second comprehensive feature matrix.

[0025] As an implementable embodiment, the center of gravity compensation model is expressed as follows:

[0026]

[0027] in, represent the second vibration data and the second strain data respectively, represent the first vibration data and the first strain data respectively, They represent the gravity compensation coefficients of vibration data and strain data, respectively. Indicates the second centroid data.

[0028] As an implementable method, the key time step of the second comprehensive feature matrix is ​​identified based on the global attention mechanism, and then the second comprehensive feature matrix is ​​weighted to obtain a third comprehensive feature matrix, including the following steps:

[0029] Calculate the importance score of each time step in the second comprehensive feature matrix by using a preset score function, and obtain the attention weight of each time step based on the importance score, wherein a higher importance score indicates a more critical time step;

[0030] Weighting the feature vector of each time step in the second comprehensive feature matrix based on the attention weight of each time step to obtain a third comprehensive feature matrix;

[0031] The preset score function is expressed as follows:

[0032]

[0033] The attention weight is expressed as follows:

[0034]

[0035] in, Indicates The importance score of each time step, represents the learnable weight matrix, Represents the second comprehensive feature matrix at time step The characteristic vector of represents the bias vector, Indicates The attention weights for each time step.

[0036] As an implementable method, the forward feature and backward feature extraction and fusion of the third comprehensive feature matrix to obtain a bidirectional time feature vector includes the following steps:

[0037] The third comprehensive feature matrix is ​​forward calculated and reverse calculated based on the bidirectional GRU in the time series to obtain the forward feature and backward feature of the current time step;

[0038] The forward feature and the backward feature are combined to obtain a bidirectional time feature vector;

[0039] Among them, the forward feature, backward feature and bidirectional time feature vector are expressed as follows:

[0040]

[0041] in, represents the bidirectional time feature vector, Represent the forward features of the current time step and the previous time step respectively, Respectively represent the backward features of the current time step and the next time step, represents the third comprehensive feature matrix, Represents a fusion operation.

[0042] As an implementable method, the weighting coefficient of each operating state data is obtained based on the change amplitude of the real-time operating state data, and each operating state integral is obtained by combining the bidirectional time feature vector and the second comprehensive feature matrix, including the following steps:

[0043] The initial weighting coefficient is adjusted based on the change amplitude of the real-time operation status data and the two-way time characteristic vector to obtain the weighting coefficient of each operation status data, which is expressed as follows:

[0044]

[0045] in, Represents the weighting coefficient of each operating status data, represents the initial value of the weighting coefficient, Indicates the change range of real-time operation status data. represents the adjustment factor, represents the bidirectional temporal eigenvector;

[0046] The weighted coefficient of each running state data is weighted with the bidirectional time feature vector, and combined with the second comprehensive feature matrix, each running state integral is obtained, which is expressed as follows:

[0047]

[0048] in, represents the integral of each operating state, Respectively represent the start and end time, represents the activation function, is a learnable weight parameter, represents the operating status data in the second comprehensive feature matrix, Represents the weighting coefficient of each operating status data, represents the bidirectional temporal feature vector.

[0049] As an implementable method, the step of judging and adjusting each running state integral, determining each running state integral weight in each fault type, and then obtaining the fault integral of each fault type includes the following steps:

[0050] Compare each running state integral with the preset integral threshold. If it exceeds the preset integral threshold, it is adjusted based on the control coefficient to determine the integral weight of each running state, which is expressed as follows:

[0051]

[0052] in, represents the integral weight of each operating state in each fault type, represents the initial weight of each operating state integral in each fault integral, The control coefficient representing the adjustment range, represents the integral of each operating state, Indicates the preset threshold value of each operating state integral;

[0053] The operating state integral and each operating state integral weight in each fault type are weighted to obtain the fault integral of each fault type.

[0054] As an implementable method, each operating state integral includes a vibration integral, a temperature integral, a pressure integral, a speed integral and a strain integral; each operating state integral weight includes a vibration integral weight, a temperature integral weight, a pressure integral weight, a speed integral weight and a strain integral weight;

[0055] The fault integral includes a drive shaft fault integral, a transmission shaft fault integral, a coupling fault integral, a clutch fault integral and a seal fault integral;

[0056] Based on the vibration integral, the speed integral, the vibration integral weight and the speed integral weight, the drive shaft fault integral and the transmission shaft fault integral are obtained;

[0057] Based on the vibration integral, the strain integral, the vibration integral weight and the strain integral weight, the coupling fault integral is obtained;

[0058] Based on the temperature integral, the speed integral, the temperature integral weight and the speed integral weight, a clutch fault integral is obtained;

[0059] Based on the temperature integral, the pressure integral, the temperature integral weight and the pressure integral weight, a seal failure integral is obtained.

[0060] As an implementable method, the drive shaft fault integral is expressed as follows:

[0061]

[0062] in, represents the drive shaft fault integral, They represent vibration integrals respectively and speed integral The corresponding vibration integral weight and speed integral weight, Represents the bidirectional time feature vector weight in drive shaft failure;

[0063] The transmission shaft fault integral is expressed as follows:

[0064]

[0065] in, represents the transmission shaft fault integral, They represent vibration integrals respectively and speed integral The corresponding vibration integral weight and speed integral weight, Represents the bidirectional time feature vector weight in drive shaft failure;

[0066] The coupling fault integral is expressed as follows:

[0067]

[0068] in, represents the coupling fault integral, They represent vibration integrals respectively and strain integral The corresponding vibration integral weight and strain integral weight, Represents the bidirectional time feature vector weight in coupling failure;

[0069] The clutch fault integral is expressed as follows:

[0070]

[0071] in, represents the clutch failure integral, Represents temperature integral and speed integral The corresponding temperature integral weight and speed integral weight, Represents the bidirectional time feature vector weight in clutch failure;

[0072] The seal failure score is expressed as follows:

[0073]

[0074] in, represents the seal failure integral, Represents temperature integral and pressure integral The corresponding temperature integral weight and pressure integral weight, Represents the bidirectional time feature vector Weight in seal failure.

[0075] As an implementable method, obtaining a fault diagnosis result according to the fault integral of each fault type includes the following steps:

[0076] Based on the fault integral of each fault type, a likelihood function of each fault type is obtained;

[0077] Based on the likelihood function of each fault type and the preset prior probabilities, the marginal probability of each fault type is obtained;

[0078] Based on the likelihood function, prior probability and marginal probability, combined with Bayesian reasoning, the posterior probability of each fault type is obtained, which is expressed as follows:

[0079]

[0080] in, Indicates that the observed fault integral Fault type under the condition The probability of occurrence, that is, the posterior probability, Indicates the fault type The likelihood function of Indicates the preset fault type The prior probability of Indicates the fault type The marginal probability of

[0081] The fault diagnosis result of the off-road forklift transmission system is obtained based on the maximum value of the posterior probability.

[0082] An artificial intelligence-based off-road forklift transmission system fault diagnosis system, comprising a data acquisition and processing module, a first comprehensive matrix calculation module, a second comprehensive matrix calculation module, a time feature calculation module, an operation state integral calculation module, a fault integral calculation and diagnosis result acquisition module;

[0083] The data acquisition and processing module is used to acquire real-time operating status data of the transmission system of the off-road forklift, pre-process the real-time operating status data, and obtain first operating status data, wherein the first operating status data at least includes first center of gravity data, first vibration data, and first strain data;

[0084] The first comprehensive matrix calculation module is used to extract multi-scale features from the first operating state data to obtain a multi-scale feature matrix and fuse the multi-scale feature matrix to obtain a first comprehensive feature matrix;

[0085] The second comprehensive matrix calculation module is used to perform gravity compensation on the first vibration data and the first strain data through the first gravity center data to update the first comprehensive feature matrix to obtain a second comprehensive feature matrix;

[0086] The temporal feature calculation module is used to identify the key time steps of the second comprehensive feature matrix based on the global attention mechanism, and then weight the second comprehensive feature matrix to obtain a third comprehensive feature matrix, extract and fuse forward features and backward features of the third comprehensive feature matrix to obtain a bidirectional temporal feature vector;

[0087] The operation status integral calculation module is used to obtain a weighting coefficient of each operation status data based on the change amplitude of the real-time operation status data, and to obtain each operation status integral by combining the bidirectional time feature vector and the second comprehensive feature matrix;

[0088] The fault score calculation and diagnosis result acquisition module is used to judge and adjust each operating state score, determine the weight of each operating state score in each fault type, and then obtain the fault score of each fault type, and obtain the fault diagnosis result according to the fault score of each fault type.

[0089] A computer program product comprises computer instructions, which, when executed by a processor, implement the following method:

[0090] Acquire real-time operating status data of the transmission system of the off-road forklift, pre-process the real-time operating status data, and obtain first operating status data, wherein the first operating status data at least includes first center of gravity data, first vibration data, and first strain data;

[0091] Perform multi-scale feature extraction on the first operating state data to obtain a multi-scale feature matrix and fuse the multi-scale feature matrix to obtain a first comprehensive feature matrix;

[0092] Performing gravity compensation on the first vibration data and the first strain data by using the first gravity center data to update the first comprehensive feature matrix to obtain a second comprehensive feature matrix;

[0093] Based on the global attention mechanism, the key time steps of the second comprehensive feature matrix are identified, and then the second comprehensive feature matrix is ​​weighted to obtain a third comprehensive feature matrix, and forward features and backward features are extracted and fused on the third comprehensive feature matrix to obtain a bidirectional time feature vector;

[0094] A weight coefficient of each operating state data is obtained based on the change amplitude of the real-time operating state data, and each operating state integral is obtained by combining the bidirectional time feature vector and the second comprehensive feature matrix;

[0095] Each operating state integral is judged and adjusted, and the weight of each operating state integral in each fault type is determined, thereby obtaining the fault integral of each fault type, and obtaining the fault diagnosis result according to the fault integral of each fault type.

[0096] An artificial intelligence-based off-road forklift transmission system fault diagnosis device comprises a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor implements the following method when executing the computer program:

[0097] Acquire real-time operating status data of the transmission system of the off-road forklift, pre-process the real-time operating status data, and obtain first operating status data, wherein the first operating status data at least includes first center of gravity data, first vibration data, and first strain data;

[0098] Perform multi-scale feature extraction on the first operating state data to obtain a multi-scale feature matrix and fuse the multi-scale feature matrix to obtain a first comprehensive feature matrix;

[0099] Performing gravity compensation on the first vibration data and the first strain data by using the first gravity center data to update the first comprehensive feature matrix to obtain a second comprehensive feature matrix;

[0100] Based on the global attention mechanism, the key time steps of the second comprehensive feature matrix are identified, and then the second comprehensive feature matrix is ​​weighted to obtain a third comprehensive feature matrix, and forward features and backward features are extracted and fused on the third comprehensive feature matrix to obtain a bidirectional time feature vector;

[0101] A weight coefficient of each operating state data is obtained based on the change amplitude of the real-time operating state data, and each operating state integral is obtained by combining the bidirectional time feature vector and the second comprehensive feature matrix;

[0102] Each operating state integral is judged and adjusted, and the weight of each operating state integral in each fault type is determined, thereby obtaining the fault integral of each fault type, and obtaining the fault diagnosis result according to the fault integral of each fault type.

[0103] The present invention has significant technical effects due to the adoption of the above technical solution:

[0104] The present application adopts artificial intelligence technologies such as multi-scale feature extraction, key time step identification, and bidirectional feature extraction, and combines with center of gravity compensation to improve the efficiency of fault diagnosis of off-road forklift transmission system. The present application performs multi-scale feature extraction on the operating state data (vibration, temperature, pressure, speed, strain and center of gravity) of the off-road forklift transmission system. The center of gravity data is used to dynamically adjust the vibration data and strain data, and then adjust the fluctuation of the off-road forklift operating state data caused by the change of the center of gravity, so as to ensure that the analysis of the operating state data can reflect the real fault situation. The features of the key time steps of the operating state data are identified and extracted, and the time-dependent features are further extracted to obtain the bidirectional time feature vector containing the deep time feature, and then the integral of each operating state data is obtained. Based on the relationship between the key components of the transmission system and the operating state data, the integral of various fault types is obtained, so as to determine the fault type of the off-road forklift transmission system and realize the fault diagnosis of the off-road forklift transmission system. The method of the present application can improve the accuracy and reliability of fault diagnosis of off-road forklift transmission system, and can significantly improve the performance of the system and the ability to process complex data, thereby reducing maintenance costs and increasing the service life of equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0105] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0106] Figure 1 It is a schematic flow diagram of the method of the present invention;

[0107] Figure 2 It is an overall schematic diagram of the system of the present invention. DETAILED DESCRIPTION

[0108] The present invention is further described in detail below in conjunction with the embodiments, which are explanations of the present invention but are not limited to the following embodiments. In the absence of conflict, the features in the following embodiments can be combined with each other.

[0109] Embodiment 1:

[0110] A fault diagnosis method for off-road forklift transmission system based on artificial intelligence, such as Figure 1 As shown, the following steps are included:

[0111] S100: acquiring real-time operating status data of a transmission system of an off-road forklift, and preprocessing the real-time operating status data to obtain first operating status data, wherein the first operating status data at least includes first center of gravity data, first vibration data, and first strain data;

[0112] S200: extracting multi-scale features from the first operating state data to obtain a multi-scale feature matrix and fusing the multi-scale feature matrix to obtain a first comprehensive feature matrix;

[0113] S300: performing gravity compensation on the first vibration data and the first strain data by using the first gravity center data to update the first comprehensive feature matrix to obtain a second comprehensive feature matrix;

[0114] S400: identifying the key time steps of the second comprehensive feature matrix based on the global attention mechanism, and then weighting the second comprehensive feature matrix to obtain a third comprehensive feature matrix, extracting and fusing forward features and backward features of the third comprehensive feature matrix to obtain a bidirectional time feature vector;

[0115] S500: obtaining a weighting coefficient of each operating state data based on the change amplitude of the real-time operating state data, and combining the bidirectional time feature vector and the second comprehensive feature matrix to obtain each operating state integral;

[0116] S600: judging and adjusting each operating state integral, determining each operating state integral weight in each fault type, thereby obtaining a fault integral of each fault type, and obtaining a fault diagnosis result according to the fault integral of each fault type.

[0117] The present application performs multi-scale feature extraction on the operating status data (vibration, temperature, pressure, speed, strain and center of gravity) of the off-road forklift transmission system. The center of gravity data is used to dynamically adjust the vibration data and strain data, and then adjust the fluctuations in the operating status data of the off-road forklift caused by the change in the center of gravity, to ensure that the analysis of the operating status data can reflect the actual fault situation. The features of the key time steps of the operating status data are identified and extracted, and the time-dependent features are further extracted to obtain a bidirectional time feature vector containing deep time features, and then the integral of each operating status data is obtained. Based on the relationship between the key components of the transmission system and the operating status data, the integral of various fault types is obtained, thereby determining the fault type of the off-road forklift transmission system and realizing fault diagnosis of the off-road forklift transmission system. The method of the present application can improve the accuracy and reliability of fault diagnosis of the off-road forklift transmission system, and can significantly improve the system's performance and ability to process complex data, thereby reducing maintenance costs and increasing equipment life.

[0118] In S100, real-time operating status data of the transmission system of the off-road forklift are obtained, and the real-time operating status data are preprocessed to obtain first operating status data, wherein the first operating status data at least includes first center of gravity data, first vibration data, and first strain data, including the following steps:

[0119] Various types of sensors installed at different positions of the off-road forklift are used to collect the operating status data of the off-road forklift's transmission system. The operating status data includes: (1) vibration data , used to monitor the vibration of the transmission shaft and coupling. The vibration sensor is installed in the middle of the transmission shaft; (2) Temperature data , used to identify overheating of the clutch, the temperature sensor is installed at the clutch; (3) pressure data , used to monitor the pressure fluctuations of the hydraulic system. The pressure sensor is installed in the main circuit of the hydraulic system (such as near the hydraulic cylinder); (4) Speed ​​data , used to track the speed changes of the transmission system. The speed sensor is installed at the output end of the drive shaft; (5) strain data , used to identify structural stress changes under off-road conditions, the strain sensor is installed on the universal joint; (6) Center of gravity data , used to monitor the changes in the center of gravity of the forklift when lifting heavy objects and performing cross-lift operations. The center of gravity sensor is installed on the forklift body and lifting device.

[0120] In the process of collecting data, different sensors will be interfered by various noises, such as thermal noise, electromagnetic interference, photon noise, etc. These noises will reduce the accuracy of the data and cause the collected data to deviate from the actual value. Therefore, it is necessary to denoise and filter the operating status data. Through denoising and filtering preprocessing, noise and external interference can be effectively reduced.

[0121] The denoised and filtered signals are normalized to ensure data consistency. The normalized operating status data will run through the entire technical solution of this application as the basic input for feature extraction, signal compensation, integral calculation and fault analysis, providing consistent and high-quality input data for subsequent feature extraction and fault diagnosis. The normalized operating status data is recorded as the first operating status data, and the first operating status data includes: the first vibration data , first temperature data , First pressure data , First Speed ​​Data , the first strain data And the first center of gravity data .

[0122] In S200, multi-scale feature extraction is performed on the first operating status data to obtain a multi-scale feature matrix and then fused to obtain a first comprehensive feature matrix.

[0123] The first operation state data can be subjected to multi-scale feature extraction through artificial intelligence technologies such as convolutional neural network (CNN) or residual network (ResNet). CNN is suitable for processing the transformed off-road forklift operation data. By increasing the depth and width of the convolution layer, the model's feature extraction capability for off-road forklift operation data can be improved. Spatial pyramid pooling introduces multiple pooling windows of different scales in the pooling layer to capture features of different sizes, thereby improving the multi-scale capability. Extending the CNN model to introduce multi-scale convolution can specifically process features of different scales and enhance the model's perception capability. ResNet solves the gradient vanishing problem in deep neural networks by introducing residual connections, thereby being able to build a deeper network structure. This deeper network structure helps to extract richer feature information and also has advantages for multi-scale feature extraction of off-road forklift operation data. Traditional signal processing methods, such as wavelet transform, can also be used to achieve multi-scale feature extraction of the first operation state data. Wavelet transform is a theory for local transformation within the time and frequency range. Through operations such as scaling and translation, wavelet transform can perform multi-scale focusing analysis on signals, providing an effective means for time-frequency analysis of non-stationary signals.

[0124] Compared with other technologies, wavelet transform has obvious advantages in processing off-road forklift operating status data: multi-resolution. With its multi-resolution characteristics, wavelet transform shows a strong ability to characterize the local characteristics of the signal in both the time domain and the frequency domain. It can gradually observe the signal from coarse to fine, so as to accurately extract useful information. Strong adaptability. The wavelet transform can select different wavelet basis functions and scale parameters according to the characteristics of the signal, so as to achieve adaptive analysis of the signal. This adaptability makes the wavelet transform have better effect when processing non-stationary and nonlinear signals. High computational efficiency. Although the wavelet transform needs to perform operations such as scaling and translation, its calculation process is relatively simple and efficient. Especially when processing large-scale data, the wavelet transform can maintain a high computational efficiency to meet the real-time requirements. Easy to implement. The algorithm of wavelet transform is relatively simple and easy to implement, which makes it widely applicable in engineering applications. Wavelet transform can also be combined with other signal processing methods (such as filtering, denoising, etc.) to further improve the processing effect.

[0125] S210: extracting multi-scale features of each running state data with different scale parameters and different time steps from each running state data based on wavelet transform with different scale parameters and different time steps in the first running state data, as shown below:

[0126]

[0127] in, Represents the wavelet coefficient, representing the first operating state data At a specific scale and specific time location The wavelet transform characteristics under the scale and time location First running status data The characteristic strength, scale parameter Used to characterize the degree of expansion and contraction of wavelet transform. The larger it is, the wider the wavelet transform stretches on the time axis, and the lower the corresponding frequency is. The smaller it is, the narrower the wavelet transform shrinks on the time axis, and the higher the corresponding frequency. Represents the time shift parameter, which controls the time position of the wavelet transform. It includes frequency domain features and time domain features. The frequency domain features characterize the amplitude and energy of the first operating status data at different frequencies, and the time domain features characterize the fluctuation and mutation of the first operating status data in different time windows. Represents the mother wavelet function, which is used for signal decomposition. Indicates the time step. In this embodiment, continuous wavelet transform is used to obtain the specific scale parameter The wavelet coefficients under the continuous wavelet transform are obtained by adjusting the scale parameters of the wavelet transform. , and obtain wavelet coefficients at different scales, so as to extract the features of multi-source fusion data in different frequency ranges.

[0128] S220: obtaining a first multi-scale feature matrix corresponding to each operating state based on the multi-scale features, fusing the first multi-scale feature matrices corresponding to each operating state to obtain a first comprehensive feature matrix, including the following steps:

[0129] (1) Extract different scales of each operating status data in the first operating status data and time step These features constitute the first multi-scale feature matrix of the running status data The rows of the first multi-scale feature matrix of each running state represent the features to be tested at the same scale (frequency) and different time steps, and each column represents the features at different scales and the same time step (position). The first multi-scale feature matrix of each running state data is The formula is as follows:

[0130]

[0131] The first multi-scale feature matrix of each operating state It includes a first multi-scale vibration characteristic matrix, a first multi-scale temperature characteristic matrix, a first multi-scale pressure characteristic matrix, a first multi-scale velocity characteristic matrix, a first multi-scale strain characteristic matrix and a first multi-scale center of gravity characteristic matrix.

[0132] (2) Fuse the first multi-scale feature matrix of each operating state to obtain the first comprehensive feature matrix , which is expressed as follows:

[0133]

[0134] in, represents the first multi-scale vibration feature matrix, represents the first multi-scale temperature feature matrix, represents the first multi-scale pressure feature matrix, represents the first multi-scale velocity feature matrix, represents the first multi-scale strain feature matrix, Represents the first multi-scale centroid feature matrix. The first multi-scale feature matrix represents the features of different operating state data at different scales (frequencies) and time steps, and all first multi-scale feature matrices together constitute the first comprehensive feature matrix.

[0135] By integrating the feature matrix, the system can capture the complex relationship between multiple operating status data, multiple scales, and multiple times, thereby improving the accuracy of fault diagnosis. This detailed multi-scale and multi-time feature extraction and fusion process ensures that the system can extract rich feature information from a single operating status data.

[0136] In S300, the first vibration data and the first strain data are subjected to gravity compensation by using the first gravity center data, and then the first comprehensive feature matrix is ​​updated to obtain the second comprehensive feature matrix, including the following steps:

[0137] S310: Low-pass filter the first center of gravity data to obtain second center of gravity data. The center of gravity data is a time-varying data that indicates the displacement of the center of gravity of the off-road forklift at different times. The data is obtained through the center of gravity sensors on the vehicle body and the lifting device. In order to eliminate unnecessary high-frequency noise and extract the main center of gravity movement trend, the center of gravity data needs to be low-pass filtered to obtain smooth second center of gravity data. The second center of gravity data characterizes the change of the center of gravity of flat turbulence.

[0138] S320: construct a gravity compensation model, and perform gravity compensation on the first vibration data and the first strain data through the second gravity data to obtain second vibration data and second strain data, wherein the gravity compensation model is expressed as follows:

[0139]

[0140] in, represent the first vibration data and the first strain data respectively; They represent the second vibration data and the second strain data respectively, which are the data after the center of gravity compensation; They represent the gravity compensation coefficient of the vibration data and the gravity compensation coefficient of the strain data respectively, and are used to adjust the influence of the gravity center change on the vibration data and the strain data. These gravity center compensation coefficients can be dynamically adjusted according to the specific operation conditions of the forklift (off-road driving and lifting height).

[0141] During off-road or heavy lifting, the vibration data will be affected by the shift in the center of gravity, for example, the vibration increases when the forklift is tilted. , which can eliminate non-fault vibrations caused by terrain changes or heavy lifting, thereby obtaining more accurate second vibration data Similarly, the strain data will fluctuate with the change of the forklift's center of gravity, especially when lifting heavy objects or off-roading. The second strain data after compensation It reflects the actual stress state more accurately and eliminates non-fault fluctuations caused by changes in the center of gravity.

[0142] The center of gravity compensation coefficient can be dynamically adjusted according to the actual operating conditions: First, by analyzing the terrain pattern of the center of gravity data (such as flat ground, ramp, gravel ground), the center of gravity compensation coefficient is adjusted so that the system can adapt to the center of gravity changes caused by different terrains. Secondly, the center of gravity compensation coefficient is automatically adjusted based on the change of lifting height to ensure that the vibration data and strain data are properly compensated during the lifting process of heavy objects to avoid misjudgment due to normal vibration or strain during the lifting process. The center of gravity compensation coefficient is adjusted comprehensively according to the terrain pattern and lifting height.

[0143] S330: Perform multi-scale feature extraction on the second vibration data and the second strain data, recalculate the multi-scale vibration feature matrix and the multi-scale strain feature matrix, and obtain a second multi-scale vibration feature matrix and a second multi-scale strain feature matrix.

[0144] S340: Use the second multi-scale vibration feature matrix and the second multi-scale strain feature matrix to update the first multi-scale vibration feature matrix and the first multi-scale strain feature matrix in the first comprehensive feature matrix to obtain a second comprehensive feature matrix.

[0145] Each running state data in the second comprehensive feature matrix is ​​the second running state data, recorded as: second vibration data , Second temperature data , Second pressure data , Second speed data , the second strain data And the second center of gravity data .

[0146] The second comprehensive feature matrix is ​​the comprehensive feature matrix after center of gravity compensation, which is used for subsequent global attention mechanism and bidirectional time feature analysis. This center of gravity compensation method ensures that the system can effectively cope with the center of gravity changes of off-road forklifts under complex working conditions during off-road and lifting operations, thereby ensuring the stability and reliability of data analysis and improving the accuracy of fault diagnosis.

[0147] The center of gravity of an off-road forklift changes greatly during operation, especially when driving on uneven terrain or lifting heavy objects. In off-road environments, such as ramps, gullies, and muddy ground, the center of gravity of the forklift will move as the terrain changes, causing the vehicle body to tilt or vibration to increase. These vibrations caused by terrain changes do not represent a transmission system failure, but they will interfere with vibration data, speed data, and strain data. In addition, changes in the vehicle body posture will also affect the pressure distribution and temperature fluctuations of the hydraulic system. If the center of gravity compensation is not performed, the diagnostic system may misjudge these normal changes caused by the terrain as fault signals. In addition, when the off-road forklift lifts heavy objects, the height and weight of the heavy objects directly affect the center of gravity position of the forklift. The center of gravity rises as the lifting height increases, increasing the risk of the vehicle overturning and causing the system load to increase or decrease. At this time, the temperature of the clutch and hydraulic system will rise, and the working pressure of the hydraulic system may also change accordingly. In addition, the movement of the center of gravity during the lifting of heavy objects will cause fluctuations in vibration data and strain data. Although these changes are normal phenomena during operation, they will interfere with the accuracy of the operating status data, so they need to be adjusted through the center of gravity compensation model to avoid misjudging faults. This application introduces a center of gravity compensation model, which uses the center of gravity data to dynamically adjust the vibration data and strain data, and then adjusts the operating status data fluctuations caused by the change of the center of gravity, ensuring that the analysis of these operating status data can reflect the actual fault situation, rather than non-fault fluctuations caused by the natural offset of the forklift's center of gravity.

[0148] In S400, the key time step of the second comprehensive feature matrix is ​​identified based on the global attention mechanism, and then the second comprehensive feature matrix is ​​weighted to obtain a third comprehensive feature matrix, and forward features and backward features are extracted and fused on the third comprehensive feature matrix to obtain a bidirectional time feature vector, including the following steps:

[0149] S410: Calculate the importance score of each time step in the second comprehensive feature matrix through a preset score function, and obtain the attention weight of each time step based on the importance score, wherein a higher importance score indicates a more critical time step.

[0150] The preset score function is expressed as follows:

[0151]

[0152] in, Indicates The importance score of each time step, represents a learnable weight matrix for linear transformation, Represents the second comprehensive feature matrix at time step The characteristic vector of Represents the bias vector.

[0153] The attention weight is expressed as follows:

[0154]

[0155] Indicates The attention weight of each time step represents the importance of the time step. The sum of the attention weights of all time steps is 1, which ensures the normalization of the attention mechanism. The time step with a higher attention weight is the key time step, which contains more important transmission system signal information. The time step with a lower attention weight is considered to be a secondary time step, which helps to extract the most important information from a large amount of data, thereby improving the performance and efficiency of the system.

[0156] S420: Attention weight based on each time step Weight the eigenvectors of each time step in the second comprehensive feature matrix to obtain the third comprehensive feature matrix , which is expressed as follows:

[0157]

[0158] The third comprehensive feature matrix is the weighted feature vector, highlighting the features of the key time steps.

[0159] S430: extracting and fusing forward features and backward features from the third comprehensive feature matrix to obtain a bidirectional time feature vector, including the following steps:

[0160] The third comprehensive feature matrix is ​​forward calculated and reverse calculated based on the bidirectional GRU in the time series to obtain the forward feature and backward feature of the current time step;

[0161] The forward feature and the backward feature are combined to obtain a bidirectional time feature vector;

[0162] Among them, the forward feature, backward feature and bidirectional time feature vector are expressed as follows:

[0163]

[0164] in, represents the bidirectional time feature vector, Represent the positive / forward features of the current time step and the previous time step respectively, Respectively represent the reverse / backward features of the current time step and the next time step, represents the third comprehensive feature matrix, Represents a merge (fusion) operation. Based on the time step information of the current time step and its previous time step. Based on the information of the current time step and the time step after it. The final output feature vector It is a bidirectional temporal feature vector that is a fusion of forward features and reverse features. It contains information from the past to the future, combines forward and backward information, represents the dependency of the time series, and is a comprehensive feature vector.

[0165] Techniques such as temporal attention mechanism or global attention mechanism or a combination of the two can be used to identify time steps in the second comprehensive feature matrix that are important for fault detection, and assign higher weights to these time steps, thereby ensuring that key features are fully utilized.

[0166] Technologies that can achieve forward feature and backward feature extraction include: backward shaving and iterative shaving, feature engineering, ARIMA model in time series analysis, state space model and bidirectional GRU model, etc. The bidirectional GRU model can effectively capture the forward and backward dependencies in the sequence data through its inherent gating mechanism and bidirectional structure, without having to try and eliminate features one by one, and usually has stronger modeling capabilities and higher prediction accuracy. This embodiment uses a bidirectional GRU model to extract forward and backward features of the third comprehensive feature matrix. The bidirectional GRU can capture the dependencies of data in time series, thereby generating a feature representation containing historical and future information.

[0167] This embodiment uses a global attention mechanism to identify and weight the features of key time steps, and then uses a bidirectional GRU model to extract the time-dependent features of the data, and finally obtains a bidirectional time feature vector containing deep time features. The bidirectional time feature vector will be used for the next calculation to ensure the accuracy and reliability of fault detection.

[0168] In S500, a weight coefficient of each operating state data is obtained based on the change amplitude of the real-time operating state data, and each operating state integral is obtained by combining the bidirectional time feature vector and the second comprehensive feature matrix, including the following steps:

[0169] S510: Adjust the initial weighting coefficient based on the change amplitude of the real-time operating status data and the two-way time characteristic vector to obtain the weighting coefficient of each operating status data. The weighting coefficient of each operating status data is used to dynamically adjust the influence of the two-way time characteristic vector on the operating status integral to ensure the sensitivity of the system to a specific operating state under different working conditions. Therefore, the weighting coefficient It is particularly important that in actual operation, the weighting coefficient needs to be dynamically adjusted based on the current working conditions (such as off-road, heavy lifting). The formula is as follows:

[0170]

[0171] in, Indicates running status data The weighting coefficient of (the weighting coefficient of each operating status data), represents the initial value of the weighting coefficient, The system is set up based on historical data and expert experience to ensure that the system can cover common working conditions, such as vibration data. It can be set to a larger value to increase the sensitivity of the transmission system to vibration anomalies; for temperature data, Can be set to a moderate value to accommodate fluctuations in clutch temperature. Indicates the change amplitude of the current real-time operating status data, which is the ratio of the increase or decrease of the current real-time operating status data compared with the operating status data of the previous time step, reflecting the volatility and abnormality of the operating status data. For example, when the amplitude of the vibration data increases abnormally, It will also increase, thereby increasing the weight. Represents the adjustment coefficient, which is used to control the sensitivity of the adjustment. The tanh function ensures that the weight changes within a certain range to avoid over-adjustment. Dynamic adjustment of the weight coefficient ensures that the weight coefficient is automatically increased when the running status data changes significantly or the bidirectional time feature vector detects an important time step.

[0172] S520: Weighting the weighted coefficient of each running state data and the bidirectional time feature vector, combined with the second comprehensive feature matrix, to obtain each running state integral, which is expressed as follows:

[0173]

[0174] in, Indicates running status data (vibration, temperature, pressure, speed, strain) operating state integration (integral for each operating state), Respectively represent the start and end time, Represents an activation function (such as ReLU), which is used to enhance the nonlinear characteristics of the running state data. is a learnable weight parameter, represents the operating state data in the second comprehensive characteristic matrix, that is, the second operating state data (operating state data after center of gravity compensation), Indicates running status data The weight coefficient is used to control the bidirectional temporal feature vector output by the bidirectional GRU Contribution to the running status integral. Running status integral It represents the accumulated eigenvalues ​​after nonlinear enhancement and weighting of the data over the entire time range.

[0175] Running status points Includes: Vibration integration , vibration signals are very important in many parts of the transmission system, especially in the fault diagnosis of drive shafts, transmission shafts and couplings. Vibration sensors are generally installed at both ends of the transmission shaft or at the connection of the clutch. Speed ​​integral The speed signal is particularly critical in the fault diagnosis of the drive shaft, transmission shaft and clutch. The speed sensor is generally installed at the output end of the differential. Pressure Integral The pressure signal is very important in the fault diagnosis of hydraulic system or differential. The pressure sensor is generally installed in the main fluid circuit of the hydraulic system. The temperature signal is very important for diagnosing faults such as clutches, differentials and seals. The temperature sensor is generally installed on the internal gear of the differential. ,If some parts of the transmission system such as the drive shaft or transmission shaft are subject to large mechanical stress, consider increasing or adjusting the weight of the strain signal. Strain sensors are generally installed on the universal joint.

[0176] Each operating status integral will serve as the basic input for subsequent fault integrals, and is used to comprehensively analyze the operating status data characteristics of different faults, further improving the accuracy of fault detection. By precisely controlling and adjusting the weighting coefficients of the operating status integrals, the system can adaptively respond to signal changes under off-road and heavy-load lifting conditions, improving the sensitivity and accuracy of fault identification. The operating status integral of this embodiment ensures that the system can accurately evaluate the importance and characteristics of operating status data under complex conditions, while providing a flexible weighting coefficient adjustment mechanism to cope with different operating conditions.

[0177] In S600, each running state integral is judged and adjusted, and each running state integral weight in each fault type is determined, thereby obtaining a fault integral of each fault type, and obtaining a fault diagnosis result according to the fault integral of each fault type, including the following steps:

[0178] S610: Compare each running state integral with a preset integral threshold. If it exceeds the preset integral threshold, adjust it based on the control coefficient to determine the weight of each running state integral in each fault type. The adaptive weight is the weight of each running state integral in a specific fault type, which is used to adjust the influence of the running state integral in the fault integral calculation. It is dynamically adjusted according to the initial setting value and the real-time running state integral to adapt to changes in actual operation, including the following steps:

[0179] (1) Set the initial adaptive weight. The initial adaptive weight of each running state integral is It is determined based on historical data and the importance of a specific fault type. For example, in a transmission shaft fault, the initial weight of the vibration integral is , initial weight of velocity integral ; In clutch failure, the initial weight of temperature integral , initial weight of velocity integral These initial weights ensure that the system responds accurately under common operating conditions and reflect the importance of each operating state integral in detecting a specific fault.

[0180] (2) Operation status integral and its corresponding preset reference maximum value (threshold) If the current running state is integrated Exceeding its threshold , the system will dynamically adjust its adaptive weight , which is expressed as follows:

[0181]

[0182] in, represents the weight of each operating state integral in each fault type, which is the weight of the adjusted operating state integral at the current time step. Indicates the set operating status The initial weight of the running status integral. A control factor that represents the adjustment range, usually set to a small value (such as 0.2) to prevent the weight from changing too much. Indicates the running status The current running status integral, Indicates the running status The reference maximum value or threshold value indicates the maximum value of the operating state under normal conditions. Exceeding its threshold When The value of will be greater than 1, thereby increasing the running status integral weight , emphasizing the role of this signal in fault detection. Does not exceed its threshold , the initial weights are kept unchanged or only slightly adjusted.

[0183] Set the initial weight value for each fault type and signal to ensure the sensitivity of the signal under normal conditions, and then dynamically adjust the importance of different sensor signals (operating state integral) in fault diagnosis according to the actual operating state integral and threshold, to ensure that the system can adaptively increase the weight of the relevant operating state when an abnormal signal is detected. And use a small adjustment coefficient To avoid instability caused by excessive weight adjustment. In this embodiment, the adaptive weights in different fault integrals can be automatically learned through model training. Through this simplified and flexible processing method of the operating status integral weight, the system can adaptively perform fault detection under different working conditions, improving the accuracy and practicality of fault integral calculation.

[0184] S620: Weight the operating status integral and the weight of each operating status integral in each fault type to obtain the fault integral of each fault type. Based on the operating status integral, the system calculates the integrals of various specific faults that may occur in the key components of the off-road forklift, and each fault integral characterizes the signal characteristics of a specific fault type. This embodiment introduces an adaptive weight mechanism in each fault integral formula to dynamically adjust the importance of different sensor signals in fault diagnosis. For different fault types, the corresponding operating status integral is selected, and the corresponding operating status integral weight is combined for weighted combination to form a fault integral for each fault type, which is used to accurately determine the fault type. The fault integrals of different fault types are expressed as follows:

[0185]

[0186] in, Indicates the fault type The fault score, Indication and fault type The operating status integral weight of the relevant operating status data, Indication and fault type The running status integral of the relevant running status data, Represents the bidirectional time feature vector The weight of .

[0187] In this embodiment, the fault score of each fault type is calculated as follows:

[0188] (1) Drive shaft failure score The fault of the drive shaft is usually manifested as distortion, breakage or bearing wear. The fault is mainly related to the vibration signal and speed signal. The drive shaft fault integral The formula is as follows:

[0189]

[0190] in, Denote the drive shaft fault integral Vibrational integrals in and speed integral The adaptive vibration integral weight and speed integral weight of the fault indicate its importance in the drive shaft fault. Represents the bidirectional time feature vector The weight of represents the weight of the time-dependent feature in the drive shaft fault analysis.

[0191] (2) Drive shaft failure score The fault of the transmission shaft is usually manifested as imbalance, bearing wear or U-joint failure. The fault is mainly related to the vibration signal and speed signal. The transmission shaft fault integral The formula is as follows:

[0192]

[0193] in, Respectively represent the vibration integral in the transmission shaft fault integral and speed integral The adaptive vibration integral weight and speed integral weight of the fault indicate its importance in the transmission shaft fault. Represents the bidirectional time feature vector The weight of represents the weight of time-dependent characteristics in transmission shaft fault analysis.

[0194] (3) Coupling failure score , coupling failure usually manifests as wear or damage, so the failure is mainly related to vibration signals and strain signals, coupling failure integral The formula is as follows:

[0195]

[0196] in, Respectively represent the vibration integral in the coupling fault integral and strain integral The adaptive vibration integral weight and strain integral weight of Represents the bidirectional time feature vector Weight in coupling failure.

[0197] (4) Clutch failure score In addition to slipping, clutch failure is usually caused by slipping. In addition to the initial release bearing, pressure plate or friction plate failure can also cause clutch failure. Therefore, clutch failure is mainly related to temperature signal and speed signal. Clutch failure integral The formula is as follows:

[0198]

[0199] in, Respectively represent the temperature integral in the clutch fault integral and speed integral Adaptive temperature integral weight and speed integral weight of Represents the bidirectional time feature vector Weight in clutch failure.

[0200] (5) Seal failure score , seal failure is mainly related to temperature signal and pressure signal, seal failure integral The formula is as follows:

[0201]

[0202] in, Respectively represent the temperature integral in the seal failure integral and pressure integral Adaptive temperature integral weight and pressure integral weight of Represents the bidirectional time feature vector Weight in seal failure.

[0203] S630: Obtaining a fault diagnosis result according to the fault score of each fault type, including the following steps:

[0204] (1) Based on the fault integral of each fault type, the likelihood function of each fault type is obtained, which is expressed as follows:

[0205]

[0206] in, Indicates that for a given fault type The fault integral is observed under the condition The probability of is the likelihood function. This embodiment assumes that the fault integral value follows a normal distribution, so Indicates the fault type The corresponding mean of the fault integral is Indicates the fault type The corresponding variance of the fault integral.

[0207] (2) Based on the likelihood function of each fault type and the preset prior probability, the marginal probability of each fault type is obtained, which is expressed as follows:

[0208]

[0209] in, Indicates the fault type The marginal probability of observing the fault integral value under all possible fault types The probability of Indicates the fault type The prior probability is pre-set based on historical data or system experience, indicating that in the absence of any observation information, the fault type Probability of occurrence, for example, if historical data indicates that the frequency of drive shaft failure is 20%, then: This processing method is to perform a weighted summation of the likelihood values ​​of all fault types and the corresponding prior probabilities to ensure that the marginal probability is calculated accurately.

[0210] (3) Based on the likelihood function, prior probability and marginal probability, combined with Bayesian reasoning, the posterior probability of each fault type is obtained, which is expressed as follows:

[0211]

[0212] in, Indicates that the observed fault integral Fault type under the condition The probability of occurrence is called the posterior probability. represents the likelihood function, Indicates the fault type The prior probability of Indicates the fault type The formula combines the likelihood function and the prior probability, and normalizes them to obtain the posterior probability of each fault type.

[0213] (4) Based on the posterior probability of the fault type, the fault type is determined and output. That is, the fault type with the largest posterior probability is the final fault type determined by the model.

[0214] Through the Bayesian reasoning process, the system can combine fault scores with historical data and statistical models to accurately calculate the probability of failure. This method ensures that the system can accurately calculate the probability of occurrence when faced with various types of faults and make diagnostic decisions based on the maximum a posteriori probability criterion.

[0215] Example 2

[0216] In this embodiment, the artificial intelligence-based off-road forklift transmission system fault diagnosis method is further explained through the steps and data in the actual implementation process.

[0217] Step 1: Data collection and preprocessing. During the operation of the off-road forklift, the sensors installed in different parts of the transmission system collect real-time operating status data such as vibration, temperature, pressure, speed, strain and center of gravity. After the collected raw data is preprocessed by filtering and normalization, the first operating status data is obtained.

[0218] Step 2: Multi-scale feature extraction. In order to capture the complex features in the data, wavelet transform is performed on the first operating state data to extract features of different scales. Wavelet transform extracts features in different frequency ranges by adjusting the scale parameters, and these features are fused into a multi-scale first comprehensive feature matrix.

[0219] Step 3: Center of gravity compensation. The preprocessed center of gravity data is low-pass filtered to obtain new center of gravity data, and the new center of gravity data is used to perform center of gravity compensation on the preprocessed vibration data and strain data to obtain new vibration data and strain data, and the first comprehensive feature matrix is ​​updated to obtain the second comprehensive feature matrix.

[0220] Step 4: Time step weight identification and bidirectional temporal feature processing: Through the global attention mechanism, the key time step features in the second comprehensive feature matrix are identified, and then the bidirectional GRU model is used to extract the time-dependent features of the signal, and finally a bidirectional temporal feature vector containing deep temporal features is obtained.

[0221] Step 5: Integrate each operating state. The weighting coefficient of each operating state data is obtained based on the change amplitude of the real-time operating state data, and weighted with the two-way time characteristic vector, and combined with the second comprehensive characteristic matrix to obtain the integral of each operating state. By accurately controlling and adjusting the weighting coefficient of the signal through the change amplitude of the real-time operating state data and the two-way time characteristic vector, the system can adaptively respond to signal changes under off-road and heavy-load lifting conditions, and improve the sensitivity and accuracy of fault identification. This detailed signal integration process ensures that the system can accurately evaluate the importance and characteristics of the signal under complex conditions, while providing a flexible weighting coefficient adjustment mechanism to cope with different operating conditions.

[0222] Step 6: Fault integration and diagnosis results. Each fault integration formula dynamically adjusts the contribution of each sensor signal in fault diagnosis through an adaptive weight mechanism. For example, the gear wear fault integration formula mainly combines the characteristics of vibration and temperature signals, while the bearing fault combines vibration, temperature and pressure signals. Through this simplified and flexible weighted coefficient calculation process, the system can adaptively perform fault detection under different working conditions, improving the accuracy and practicality of the fault integration calculation. As shown in Table 1, the initial weight, preset threshold and control coefficient of each operating state integral in the fault integration are designed.

[0223] Table 1 Design of initial weight, preset threshold and control coefficient for each operating state integral in fault integral

[0224]

[0225] The fault diagnosis result is obtained according to the fault integral of each fault type. Using Bayesian reasoning, based on the fault integral of each fault type and the preset prior probabilities, the likelihood function and marginal probability of each fault type are obtained, and then the posterior probability of each fault type is obtained. By comparing the posterior probability values ​​of each fault type, the system determines the most likely fault type and outputs the diagnosis result. Based on the fault diagnosis results, the system outputs the specific fault type of the off-road forklift transmission system and provides maintenance personnel with corresponding predictive maintenance suggestions, such as timely replacement of lubricating oil, checking the status of gears or bearings, etc. Table 2 shows the parameter values ​​in the Bayesian reasoning process.

[0226] Table 2 Parameter values ​​in the Bayesian inference process

[0227]

[0228] The fault type with the largest posterior probability is taken as the final fault type, and the fault of the off-road forklift transmission system is a drive shaft fault.

[0229] Embodiment 3:

[0230] An artificial intelligence-based fault diagnosis system for off-road forklift transmission system, such as Figure 2 As shown, it includes a data acquisition and processing module 100, a first comprehensive matrix calculation module 200, a second comprehensive matrix calculation module 300, a time feature calculation module 400, an operation status integral calculation module 500, and a fault integral calculation and diagnosis result acquisition module 600;

[0231] The data acquisition and processing module 100 is used to acquire real-time operating status data of the transmission system of the off-road forklift, pre-process the real-time operating status data, and obtain first operating status data, wherein the first operating status data at least includes first center of gravity data, first vibration data, and first strain data;

[0232] The first comprehensive matrix calculation module 200 is used to extract multi-scale features from the first operating state data to obtain a multi-scale feature matrix and fuse it to obtain a first comprehensive feature matrix;

[0233] The second comprehensive matrix calculation module 300 is used to perform gravity compensation on the first vibration data and the first strain data through the first gravity center data to update the first comprehensive feature matrix to obtain a second comprehensive feature matrix;

[0234] The time feature calculation module 400 is used to identify the key time steps of the second comprehensive feature matrix based on the global attention mechanism, and then weight the second comprehensive feature matrix to obtain a third comprehensive feature matrix, extract and fuse forward features and backward features of the third comprehensive feature matrix, and obtain a bidirectional time feature vector;

[0235] The operating state integral calculation module 500 is used to obtain a weighting coefficient of each operating state data based on the change amplitude of the real-time operating state data, and to obtain each operating state integral by combining the bidirectional time feature vector and the second comprehensive feature matrix;

[0236] The fault score calculation and diagnosis result acquisition module 600 is used to judge and adjust each operating state score, determine the weight of each operating state score in each fault type, and then obtain the fault score of each fault type, and obtain the fault diagnosis result according to the fault score of each fault type.

[0237] Various changes and modifications can be made without departing from the spirit and scope of the present invention, and all equivalent technical solutions also belong to the scope of the present invention.

[0238] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0239] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, devices, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0240] The present invention is described with reference to the flowcharts and / or block diagrams of the method, terminal device (system), and computer program product according to the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0241] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device, which implements the process Figure 1A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0242] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device so that a series of operating steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable terminal device to implement the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0243] It should be noted that:

[0244] The "one embodiment" or "embodiment" mentioned in the specification means that the specific features, structures or characteristics described in conjunction with the embodiment are included in at least one embodiment of the present invention. Therefore, the phrases "one embodiment" or "embodiment" appearing in various places throughout the specification do not necessarily refer to the same embodiment.

[0245] In addition, it should be noted that the shapes and names of the parts and components of the specific embodiments described in this specification may be different. Any equivalent or simple changes made based on the structure, features and principles described in the patent concept of the present invention are included in the protection scope of the patent of the present invention. The technicians in the technical field of the present invention can make various modifications or supplements to the specific embodiments described or replace them in a similar manner, as long as they do not deviate from the structure of the present invention or exceed the scope defined by the claims, they should all fall within the protection scope of the present invention.

Claims

1. A fault diagnosis method for off-road forklift transmission system based on artificial intelligence, characterized in that: The following steps are involved: Acquire real-time operating status data of the transmission system of the off-road forklift, pre-process the real-time operating status data, and obtain first operating status data, wherein the first operating status data at least includes first center of gravity data, first vibration data, and first strain data; Perform multi-scale feature extraction on the first operating state data to obtain a multi-scale feature matrix and fuse the multi-scale feature matrix to obtain a first comprehensive feature matrix; Performing gravity compensation on the first vibration data and the first strain data by using the first gravity center data to update the first comprehensive feature matrix to obtain a second comprehensive feature matrix; Based on the global attention mechanism, the key time steps of the second comprehensive feature matrix are identified, and then the second comprehensive feature matrix is ​​weighted to obtain a third comprehensive feature matrix, and forward features and backward features are extracted and fused on the third comprehensive feature matrix to obtain a bidirectional time feature vector; Based on the change amplitude of the real-time operation status data, a weight coefficient of each operation status data is obtained, and each operation status integral is obtained by combining the bidirectional time feature vector and the second comprehensive feature matrix; Each operating state integral is judged and adjusted, and the weight of each operating state integral in each fault type is determined, thereby obtaining the fault integral of each fault type, and obtaining the fault diagnosis result according to the fault integral of each fault type.

2. The artificial intelligence-based off-road forklift transmission system fault diagnosis method according to claim 1 is characterized in that: The step of extracting multi-scale features from the first operating state data to obtain a multi-scale feature matrix and fusing the multi-scale feature matrix to obtain a first comprehensive feature matrix includes the following steps: Extracting multi-scale features of each running state data with different scale parameters and different time steps from each running state data in the first running state data by wavelet transform based on different scale parameters and different time steps; Based on the multi-scale features, a first multi-scale feature matrix corresponding to each operating state is obtained, and the first multi-scale feature matrix corresponding to each operating state is fused to obtain a first comprehensive feature matrix; Among them, wavelet transform is expressed as follows: in, represents the wavelet transform coefficients, represents the scale parameter, represents the time shift parameter, represents the first running status data, represents the mother wavelet function, Represents the time step.

3. The fault diagnosis method for off-road forklift transmission system based on artificial intelligence according to claim 1 is characterized in that: The method of performing gravity compensation on the first vibration data and the first strain data by using the first gravity center data to update the first comprehensive feature matrix to obtain the second comprehensive feature matrix includes the following steps: Performing low-pass filtering on the first centroid data to obtain second centroid data; Constructing a gravity compensation model, performing gravity compensation on the first vibration data and the first strain data through the second gravity data, and obtaining the second vibration data and the second strain data; Performing multi-scale feature extraction on the second vibration data and the second strain data to obtain a second multi-scale vibration feature matrix and a second multi-scale strain feature matrix; The first comprehensive feature matrix is ​​updated by using the second multi-scale vibration feature matrix and the second multi-scale strain feature matrix to obtain a second comprehensive feature matrix.

4. The artificial intelligence-based off-road forklift transmission system fault diagnosis method according to claim 3 is characterized in that: The center of gravity compensation model is expressed as follows: in, represent the second vibration data and the second strain data respectively, represent the first vibration data and the first strain data respectively, They represent the gravity compensation coefficients of vibration data and strain data, respectively. Indicates the second centroid data.

5. The method for fault diagnosis of transmission system of off-road forklift based on artificial intelligence according to claim 1, characterized in that: The method of identifying the key time steps of the second comprehensive feature matrix based on the global attention mechanism and then weighting the second comprehensive feature matrix to obtain the third comprehensive feature matrix includes the following steps: Calculate the importance score of each time step in the second comprehensive feature matrix by using a preset score function, and obtain the attention weight of each time step based on the importance score, wherein a higher importance score indicates a more critical time step; Weighting the feature vector of each time step in the second comprehensive feature matrix based on the attention weight of each time step to obtain a third comprehensive feature matrix; The preset score function is expressed as follows: The attention weight is expressed as follows: in, Indicates The importance score of each time step, represents the learnable weight matrix, Represents the second comprehensive feature matrix at time step The characteristic vector of represents the bias vector, Indicates The attention weights for each time step.

6. The method for fault diagnosis of transmission system of off-road forklift based on artificial intelligence according to claim 1, characterized in that: The extracting and fusing forward features and backward features of the third comprehensive feature matrix to obtain a bidirectional time feature vector comprises the following steps: The third comprehensive feature matrix is ​​forward calculated and reverse calculated based on the bidirectional GRU in the time series to obtain the forward feature and backward feature of the current time step; The forward feature and the backward feature are combined to obtain a bidirectional time feature vector; Among them, the forward feature, backward feature and bidirectional time feature vector are expressed as follows: in, Represent the forward features of the current time step and the previous time step respectively, Respectively represent the backward features of the current time step and the next time step, represents the third comprehensive feature matrix, represents the bidirectional time feature vector, Represents a fusion operation.

7. The method for fault diagnosis of transmission system of off-road forklift based on artificial intelligence according to claim 1, characterized in that: The weighting coefficient of each running state data is obtained based on the change amplitude of the real-time running state data, and each running state integral is obtained by combining the bidirectional time feature vector and the second comprehensive feature matrix, including the following steps: The initial weighting coefficient is adjusted based on the change amplitude of the real-time operation status data and the two-way time characteristic vector to obtain the weighting coefficient of each operation status data, which is expressed as follows: in, Represents the weighting coefficient of each operating status data, represents the initial value of the weighting coefficient, Indicates the change range of real-time operation status data. represents the adjustment factor, represents the bidirectional temporal eigenvector; The weighted coefficient of each running state data is weighted with the bidirectional time feature vector, and combined with the second comprehensive feature matrix, each running state integral is obtained, which is expressed as follows: in, represents the integral of each operating state, Respectively represent the start and end time, represents the activation function, is a learnable weight parameter, represents the operating status data in the second comprehensive feature matrix, Represents the weighting coefficient of each operating status data, represents the bidirectional temporal feature vector.

8. The method for fault diagnosis of transmission system of off-road forklift based on artificial intelligence according to claim 1, characterized in that: The step of judging and adjusting each running state integral, determining each running state integral weight in each fault type, and then obtaining the fault integral of each fault type includes the following steps: Compare each running state integral with the preset integral threshold. If it exceeds the preset integral threshold, it is adjusted based on the control coefficient to determine the integral weight of each running state, which is expressed as follows: in, represents the integral weight of each operating state in each fault type, represents the initial weight of each operating state integral in each fault integral, The control coefficient representing the adjustment range, represents the integral of each operating state, Indicates the preset threshold value of each operating state integral; The operating state integral and each operating state integral weight in each fault type are weighted to obtain the fault integral of each fault type.

9. The method for fault diagnosis of transmission system of off-road forklift based on artificial intelligence according to claim 1, characterized in that: Each operating state integral includes vibration integral, temperature integral, pressure integral, velocity integral and strain integral; each operating state integral weight includes vibration integral weight, temperature integral weight, pressure integral weight, velocity integral weight and strain integral weight; The fault integral includes a drive shaft fault integral, a transmission shaft fault integral, a coupling fault integral, a clutch fault integral and a seal fault integral; Based on the vibration integral, the speed integral, the vibration integral weight and the speed integral weight, the drive shaft fault integral and the transmission shaft fault integral are obtained; Based on the vibration integral, the strain integral, the vibration integral weight and the strain integral weight, the coupling fault integral is obtained; Based on the temperature integral, the speed integral, the temperature integral weight and the speed integral weight, a clutch fault integral is obtained; Based on the temperature integral, the pressure integral, the temperature integral weight and the pressure integral weight, a seal failure integral is obtained.

10. The method for fault diagnosis of transmission system of off-road forklift based on artificial intelligence according to claim 9, characterized in that: The drive shaft fault integral is expressed as follows: in, represents the drive shaft fault integral, They represent vibration integrals respectively and speed integral The corresponding vibration integral weight and speed integral weight, Represents the bidirectional time feature vector weight in drive shaft failure; The transmission shaft fault integral is expressed as follows: in, represents the transmission shaft fault integral, Represent the vibration integral and speed integral The corresponding vibration integral weight and speed integral weight, Represents the bidirectional time feature vector weight in drive shaft failure; The coupling fault integral is expressed as follows: in, represents the coupling fault integral, Represent the vibration integral and strain integral The corresponding vibration integral weight and strain integral weight, Represents the bidirectional time feature vector weight in coupling failure; The clutch fault integral is expressed as follows: in, represents the clutch failure integral, Represents temperature integral and speed integral The corresponding temperature integral weight and speed integral weight, Represents the bidirectional time feature vector weight in clutch failure; The seal failure integral is expressed as follows: in, represents the seal failure integral, Represents the temperature integral and pressure integral The corresponding temperature integral weight and pressure integral weight, Represents the bidirectional time feature vector Weight in seal failure.

11. The method for fault diagnosis of transmission system of off-road forklift based on artificial intelligence according to claim 1, characterized in that: Obtaining a fault diagnosis result according to the fault integral of each fault type comprises the following steps: Based on the fault integral of each fault type, a likelihood function of each fault type is obtained; Based on the likelihood function of each fault type and the preset prior probabilities, the marginal probability of each fault type is obtained; Based on the likelihood function, prior probability and marginal probability, combined with Bayesian reasoning, the posterior probability of each fault type is obtained, which is expressed as follows: in, Indicates that the observed fault integral Fault type under the condition The probability of occurrence, that is, the posterior probability, Indicates the fault type The likelihood function of Indicates the preset fault type The prior probability of Indicates the fault type The marginal probability of The fault diagnosis result of the off-road forklift transmission system is obtained based on the maximum value of the posterior probability.

12. An artificial intelligence-based off-road forklift transmission system fault diagnosis system, characterized in that: It includes a data acquisition and processing module, a first comprehensive matrix calculation module, a second comprehensive matrix calculation module, a time feature calculation module, an operation status integral calculation module, and a fault integral calculation and diagnosis result acquisition module; The data acquisition and processing module is used to acquire real-time operating status data of the transmission system of the off-road forklift, pre-process the real-time operating status data, and obtain first operating status data, wherein the first operating status data at least includes first center of gravity data, first vibration data, and first strain data; The first comprehensive matrix calculation module is used to extract multi-scale features from the first operating state data to obtain a multi-scale feature matrix and fuse the multi-scale feature matrix to obtain a first comprehensive feature matrix; The second comprehensive matrix calculation module is used to perform gravity compensation on the first vibration data and the first strain data through the first gravity center data to update the first comprehensive feature matrix to obtain a second comprehensive feature matrix; The temporal feature calculation module is used to identify the key time steps of the second comprehensive feature matrix based on the global attention mechanism, and then weight the second comprehensive feature matrix to obtain a third comprehensive feature matrix, extract and fuse forward features and backward features of the third comprehensive feature matrix to obtain a bidirectional temporal feature vector; The operation status integral calculation module is used to obtain a weighting coefficient of each operation status data based on the change amplitude of the real-time operation status data, and to obtain each operation status integral by combining the bidirectional time feature vector and the second comprehensive feature matrix; The fault score calculation and diagnosis result acquisition module is used to judge and adjust each operating state score, determine the weight of each operating state score in each fault type, and then obtain the fault score of each fault type, and obtain the fault diagnosis result according to the fault score of each fault type.

13. A computer program product, characterized in that The method comprises computer instructions, which, when executed by a processor, implement the method according to any one of claims 1 to 11.

14. An artificial intelligence-based off-road forklift transmission system fault diagnosis device, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 11 is implemented.

Images