A method and system for diagnosing a planetary gearbox failure

Abstract

The application provides a kind of planetary gearbox fault diagnosis method and system, belong to the technical field of cross-condition fault diagnosis, method includes to vibration acceleration signal is preprocessed, to obtain pulse sequence data set;Pulse sequence data set is extracted to space-time feature by convolution pulse neural network feature extraction module, to obtain pulse output;Determine the pulse firing rate of single neuron based on pulse output, determine continuous feature representation based on pulse firing rate;Based on continuous feature representation, construct double-branch processing structure, based on double-branch processing structure, construct total loss function, based on total loss function and back propagation algorithm, update convolution pulse neural network;The vibration acceleration signal of current to be diagnosed target domain is input into updated convolution pulse neural network for fault diagnosis, to output fault diagnosis result, the application can realize the stable extraction and accurate identification of fault feature under cross-condition.

Description

Technical Field

[0001] This invention belongs to the technical field of cross-operating condition fault diagnosis, and specifically relates to a method and system for diagnosing planetary gearbox faults. Background Technology

[0002] Planetary gearboxes, as key components in complex mechanical transmission systems, provide high torque and high-precision power output within limited space and are widely used in wind power generation, aerospace, and industrial transmission. Failures, such as broken gear teeth, pitting, or wear, can not only reduce equipment performance but also, in severe cases, lead to major safety accidents. Therefore, identifying the operating status of planetary gearboxes is crucial for equipment stability and industrial production safety.

[0003] Traditional neural networks are typically based on continuous value calculation models, which have limitations when processing vibration signals that are impactful, sparsity-dependent, and highly time-dependent. In cross-condition or cross-equipment scenarios, model performance often degrades significantly due to differences in data distribution. To address this distributional discrepancy issue, existing transfer learning methods generally rely on the Gaussian distribution assumption, using mean and covariance alignment to achieve cross-domain transfer. However, in practical vibration signal analysis, especially when using spiking neural networks (SNNs) for modeling, their output features typically exhibit sparse, discrete pulse-like characteristics, making it difficult to satisfy the Gaussian distribution assumption and thus limiting the effectiveness of the aforementioned methods.

[0004] On the other hand, spiking neural networks, as an information processing model closer to the biological nervous system, possess event-driven, low-power, and excellent temporal information modeling capabilities, and have shown potential advantages in the field of time-series signal processing in recent years. In particular, convolutional spiking neural networks (CSNNs) can simultaneously extract spatial and temporal features, making them suitable for modeling transient impact features in vibration signals. However, existing technologies still lack methods for distribution modeling and cross-condition alignment of CSNN output features, and a mature and effective technical solution has not yet been developed. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention proposes a planetary gearbox fault diagnosis method and system. This method addresses the shortcomings of existing cross-condition fault diagnosis technologies, such as the difficulty of effectively characterizing the impact and sparse temporal features of vibration signals using traditional deep learning methods based on continuous value calculations, the difficulty of statistical modeling and cross-domain distribution alignment of discrete output features of pulse neural networks, and the fact that existing transfer learning methods rely on continuous feature spaces and specific distribution assumptions and are not applicable to pulse feature scenarios.

[0006] In a first aspect, the present invention provides the following technical solution: a method for diagnosing planetary gearbox faults, comprising: The vibration acceleration signal of the planetary gearbox during operation is acquired, and the vibration acceleration signal is preprocessed to obtain a pulse sequence dataset; A convolutional spiking neural network feature extraction module is constructed to extract spatiotemporal features from the pulse sequence dataset to obtain pulse output. The pulse firing rate of a single neuron is determined based on the pulse output, and the continuous feature representation is determined based on the pulse firing rate. A dual-branch processing structure is constructed based on the continuous feature representation, a total loss function is constructed based on the dual-branch processing structure, and the convolutional spiking neural network is updated based on the total loss function and the backpropagation algorithm. The vibration acceleration signal of the target domain to be diagnosed is obtained, and the vibration acceleration signal of the target domain to be diagnosed is input into the updated convolutional pulse neural network for fault diagnosis, so as to output the fault diagnosis result.

[0007] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention proposes to use convolutional spiking neural networks to model vibration signals, making full use of the advantages of spiking neurons in time-series modeling and energy efficiency to achieve highly sensitive capture of impact-type fault features. Considering that the pulse features are highly discrete and difficult to perform statistical analysis, a pulse feature continuity module is used to transform the time-series pulse response into stable continuous features. A fusion distribution alignment mechanism based on a dual-branch structure is constructed. By introducing pseudo-labels, fine alignment of the source domain and the target domain at the category level is achieved. At the same time, the reliability of pseudo-labels is improved by combining confidence screening and dynamic update strategies. Furthermore, the classification and alignment process is optimized in a coordinated manner through a fusion loss function, which will improve the fault identification accuracy and generalization ability of planetary gearboxes under complex and variable working conditions. Replacing traditional convolutional neural networks with convolutional spiking neural networks (CNNs) can more effectively acquire the impact and transient features of vibration signals, making it particularly suitable for identifying non-stationary signals such as gear meshing faults. A pulse feature continuity module is employed to transform pulse firing behavior into a stable, continuous feature representation, thereby effectively integrating CNNs with traditional statistical modeling and machine learning methods. A dual-branch structure with a pseudo-label branch and a distribution alignment branch is designed to decouple the target domain pseudo-label generation from the feature alignment process, avoiding the mutual interference between label noise and the alignment process in traditional methods, thus improving overall training stability and reliability. By setting a confidence screening strategy for target domain samples, pseudo-labels are dynamically updated during training, effectively reducing the risk of erroneous label propagation and improving the stability of model training and final recognition accuracy.

[0008] Preferably, the step of preprocessing the vibration acceleration signal to obtain a pulse sequence dataset includes: The vibration acceleration signal is divided into several sub-samples using a sliding window. Each sub-sample is normalized, and the normalized sub-samples are mapped to pulse firing frequencies using firing rate encoding to obtain a pulse sequence dataset.

[0009] Preferably, the convolutional spiking neural network feature extraction module adopts a serial cascaded and functionally layered design. The convolutional spiking neural network feature extraction module includes two convolutional modules, each of which is composed of a convolutional layer, a batch normalization layer, and a leaky integral ignition layer connected in series. The dynamic changes in neuronal membrane potential and the pulse ignition process of the leakage integral ignition layer are described as follows: ; In the formula, for Membrane potential at time t, The membrane time constant is This is the resting potential. for Input current at any given moment; When membrane potential Exceeding the preset ignition threshold At this time, the neuron will generate a pulse signal, which will be transmitted through the step function. After encoding and outputting the pulses, the pulse outputs of the corresponding neurons in the corresponding layer of the convolutional spiking neural network are obtained: ; In the formula, For the first convolutional spiking neural network Layer One neuron in The pulse output at any given moment.

[0010] Preferably, the step of determining the pulse firing rate of a single neuron based on the pulse output, and determining the continuous feature based on the pulse firing rate, includes: The pulse output is combined into multiple channels to obtain a multi-channel discrete pulse sequence, and a preset time window is determined. Calculate the spiking rate of a single neuron within a preset time window: ; In the formula, For the first convolutional spiking neural network Layer One neuron in Pulse output at any moment For the first convolutional spiking neural network Layer The firing rate of each neuron; The pulse firing rate is standardized, and the standardized pulse firing rates are combined in channel order to obtain a continuous feature vector. : ; In the formula, For the standardized convolutional spiking neural network, the first... Layer The firing rate of each neuron; The continuous feature vector Nonlinear mapping is performed in the input fully connected layer to obtain continuous feature representations. : ; In the formula, It is a non-linear activation function. These are learnable weights and learnable biases.

[0011] Preferably, the dual-branch processing structure includes a pseudo-label branch and a feature distribution alignment branch; The pseudo-label generation branch is used to utilize unlabeled target domain data, input target domain features into a multi-layer fully connected network classifier, connect a Softmax function at the end, map the output to class probabilities in the 0-1 interval with a sum of 1, select samples with confidence scores higher than a set threshold as reliable samples based on the probabilities, and assign corresponding pseudo-labels. The target domain samples with pseudo-labels are then retrained to enhance the model's ability to discriminate target domain data. The feature distribution alignment branch is used to reduce the distribution difference between the source and target domains. It inputs source and target domain samples into the feature extraction network to obtain the corresponding feature representations. and and through the distribution metric function right , Constraints are applied to reduce the edge distribution differences between the source and target domains, including distribution alignment loss. for: .

[0012] Preferably, the total loss function for: ; In the formula, For pseudo-label loss, To balance the hyperparameters.

[0013] Secondly, the present invention provides the following technical solution: a planetary gearbox fault diagnosis system, the system comprising: The preprocessing module is used to acquire the vibration acceleration signal of the planetary gearbox during operation, and to preprocess the vibration acceleration signal to obtain a pulse sequence dataset. An extraction module is used to construct a convolutional spiking neural network feature extraction module, which extracts spatiotemporal features from the pulse sequence dataset to obtain pulse output. The pulse module is used to determine the pulse firing rate of a single neuron based on the pulse output, and to determine the continuous feature representation based on the pulse firing rate; The branching module is used to construct a dual-branch processing structure based on the continuous feature representation, construct a total loss function based on the dual-branch processing structure, and update the convolutional spiking neural network based on the total loss function and the backpropagation algorithm. The diagnostic module is used to acquire the vibration acceleration signal of the target domain to be diagnosed, input the vibration acceleration signal of the target domain to be diagnosed into the updated convolutional pulse neural network for fault diagnosis, and output the fault diagnosis result.

[0014] Preferably, the preprocessing module is used for: The vibration acceleration signal is divided into several sub-samples using a sliding window. Each sub-sample is normalized, and the normalized sub-samples are mapped to pulse firing frequencies using firing rate encoding to obtain a pulse sequence dataset.

[0015] Thirdly, the present invention provides the following technical solution: a computer, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the planetary gearbox fault diagnosis method as described above.

[0016] Fourthly, the present invention provides the following technical solution: a storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the planetary gearbox fault diagnosis method as described above. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a planetary gearbox fault diagnosis method provided in Embodiment 1 of the present invention; Figure 2 This is a structural block diagram of the planetary gearbox fault diagnosis system provided in Embodiment 2 of the present invention; Figure 3 This is a schematic diagram of the hardware structure of a computer provided for another embodiment of the present invention.

[0019] The embodiments of the present invention will be further described below with reference to the accompanying drawings. Detailed Implementation

[0020] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain embodiments of the present invention, and should not be construed as limiting the present invention.

[0021] Example 1 In Embodiment 1 of the present invention, as Figure 1 As shown, a planetary gearbox fault diagnosis method includes: S1. Obtain the vibration acceleration signal of the planetary gearbox during operation, and preprocess the vibration acceleration signal to obtain a pulse sequence dataset; Specifically, a planetary gearbox fault simulation experimental platform was built, including accelerometers, a NIPXI-1042 data acquisition card, a three-phase asynchronous motor, and a PC. The motor drives the gearbox to operate stably. Accelerometers were placed on the outer shell, input shaft end, and output shaft end of the planetary gearbox, preferably in different directions, to obtain vibration response information at multiple locations. The vibration signals collected by the sensors were continuously sampled by the data acquisition system, and operating parameters such as speed and load were recorded simultaneously. The sampling frequency and sampling duration were set, and the collected data was initially screened to remove obvious abnormal interference signals. The vibration acceleration signals containing sufficient impact characteristics and timing information were selected as the original signal dataset of the planetary gearbox.

[0022] The preprocessing process includes: The vibration acceleration signal is divided into several sub-samples using a sliding window. Each sub-sample is normalized, and the normalized sub-samples are mapped to pulse firing frequencies using firing rate encoding to obtain a pulse sequence dataset.

[0023] Specifically, the acquired planetary gearbox vibration signals undergo preprocessing, including signal segmentation, normalization, and sample construction. The original vibration signals are divided into sliding window segments of fixed length to form a sample dataset. Each sample is then normalized, mapping its amplitude to a preset range [0,1] to eliminate the dimensional effects caused by differences in acquisition conditions and operating conditions, ensuring consistency in numerical scale among the samples. Based on the normalized signal amplitudes, a firing rate mapping relationship, i.e., firing rate encoding, is constructed, ensuring that the signal amplitude corresponds to the pulse firing frequency. This allows signals with larger amplitudes to generate more pulses per unit time, while signals with smaller amplitudes generate fewer pulses, thus achieving the conversion of continuous signals to pulse representation and obtaining a pulse sequence dataset.

[0024] S2. Construct a convolutional spiking neural network feature extraction module, and extract spatiotemporal features from the pulse sequence dataset through the convolutional spiking neural network feature extraction module to obtain pulse output; Specifically, the convolutional spiking neural network feature extraction module adopts a serial cascaded and functionally layered design. The convolutional spiking neural network feature extraction module includes two convolutional modules, each of which is composed of a convolutional layer, a batch normalization layer, and a leaky integral ignition layer connected in series. The dynamic changes in neuronal membrane potential and the pulse ignition process of the leakage integral ignition layer are described as follows: ; In the formula, for Membrane potential at time t, The membrane time constant is This is the resting potential. for Input current at any given moment; When membrane potential Exceeding the preset ignition threshold At this time, the neuron will generate a pulse signal, which will be transmitted through the step function. After encoding and outputting the pulses, the pulse outputs of the corresponding neurons in the corresponding layer of the convolutional spiking neural network are obtained: ; In the formula, For the first convolutional spiking neural network Layer One neuron in Pulse output at any given moment; Specifically, the neuron membrane potential is jointly regulated by the membrane time constant, resting potential, and real-time input current. When the accumulated membrane potential exceeds the preset ignition threshold, the neuron will generate a pulse signal, which is encoded and output through a step function to obtain the discrete pulse result of the corresponding neuron in the corresponding layer of the convolutional spiking neural network.

[0025] S3. Determine the pulse firing rate of a single neuron based on the pulse output, and determine the continuous feature representation based on the pulse firing rate; Specifically, based on the requirements of modeling and distribution alignment, the discrete pulse features output by the convolutional spiking neural network are processed into a continuous form, so that the resulting features retain the pulse temporal statistical characteristics while having good distribution alignment ability, thereby avoiding the problem of temporal information loss in traditional continuous forms.

[0026] Step S3 includes: S31. Combine the pulse output into multiple channels to obtain a multi-channel discrete pulse sequence, and determine a preset time window. Calculate the spiking rate of a single neuron within a preset time window: ; In the formula, For the first convolutional spiking neural network Layer One neuron in Pulse output at any moment For the first convolutional spiking neural network Layer The firing rate of each neuron; S32. Standardize the pulse firing rate, and combine the standardized pulse firing rates according to channel order to obtain a continuous feature vector. : ; In the formula, For the standardized convolutional spiking neural network, the first... Layer The firing rate of each neuron; The standardization process is as follows: ; In the formula, These are the characteristic mean and standard deviation, respectively. To prevent tiny constants with a denominator of zero.

[0027] S33, the continuous feature vector Nonlinear mapping is performed in the input fully connected layer to obtain continuous feature representations. : ; In the formula, It is a non-linear activation function. These are learnable weights and learnable biases.

[0028] Meanwhile, the above processing achieves smoothing and structured expression of the output features of the spiking neural network, transforming the original discrete pulse sequence into a stable continuous feature representation, providing a foundation for statistical distribution modeling of cross-condition data and subsequent distribution alignment.

[0029] S4. Construct a dual-branch processing structure based on the continuous feature representation, construct a total loss function based on the dual-branch processing structure, and update the convolutional spiking neural network based on the total loss function and the backpropagation algorithm; The dual-branch processing structure includes a pseudo-label branch and a feature distribution alignment branch. The pseudo-label generation branch is used to utilize unlabeled target domain data, input target domain features into a multi-layer fully connected network classifier, connect a Softmax function at the end, map the output to class probabilities in the 0-1 interval with a sum of 1, select samples with confidence scores higher than a set threshold as reliable samples based on the probabilities, and assign corresponding pseudo-labels. The target domain samples with pseudo-labels are then retrained to enhance the model's ability to discriminate target domain data. The feature distribution alignment branch is used to reduce the distribution difference between the source and target domains. It inputs source and target domain samples into the feature extraction network to obtain the corresponding feature representations. and and through the distribution metric function right , Constraints are applied to reduce the edge distribution differences between the source and target domains, including distribution alignment loss. for: .

[0030] During training, the distribution alignment loss and pseudo-label loss are weighted and fused to obtain the total loss function: ; In the formula, For pseudo-label loss, To balance the hyperparameters; By optimizing the dual-branch structure, the model can reduce inter-domain differences while improving its ability to identify fault characteristics in the target domain, thereby achieving more accurate cross-condition fault diagnosis of planetary gearboxes.

[0031] Specifically, based on the total loss function, the feature extraction layer, feature continuity module, and classifier parameters of the convolutional spiking neural network are synchronously iteratively updated. In each training round, samples from the source domain and the target domain are simultaneously input into the model. Feature representations and prediction results are obtained through forward propagation. The classification error and cross-domain distribution differences are comprehensively constrained based on the total loss function. The parameters are synchronously optimized through backpropagation. As the model parameters are continuously updated, the pseudo-labels of the target domain are dynamically updated, thereby gradually improving the accuracy and stability of distribution alignment. Ultimately, the model achieves stable convergence and high-precision fault identification under complex working conditions.

[0032] S5. Obtain the vibration acceleration signal of the target domain to be diagnosed, and input the vibration acceleration signal of the target domain to be diagnosed into the updated convolutional pulse neural network for fault diagnosis, so as to output the fault diagnosis result. Specifically, the updated convolutional spiking neural network's feature extraction network encodes the input signal, which is then converted into a stable feature representation by a continuous transformation module before being input into a classifier. The classifier outputs the corresponding fault category probability distribution. Based on the maximum probability criterion, the fault type of the current sample is determined, thus enabling the identification of the planetary gearbox's operating status. Simultaneously, the confidence level is used to assess the reliability of the identification results. When the confidence level falls below a preset threshold, the state is determined to be abnormal or unknown, thereby achieving accurate identification of known faults and effective early warning of potential new faults.

[0033] The planetary gearbox fault diagnosis method provided in Embodiment 1 of this invention proposes to model vibration signals using a convolutional spiking neural network. This fully utilizes the advantages of spiking neurons in time-series modeling and energy efficiency to achieve highly sensitive capture of impact-type fault features. Considering the strong discreteness of spiking features and the difficulty in statistical analysis, a spiking feature continuity module is used to transform the time-series spiking response into stable continuous features. A fusion distribution alignment mechanism based on a dual-branch structure is constructed. By introducing pseudo-labels, fine alignment of the source and target domains at the category level is achieved. At the same time, confidence screening and dynamic update strategies are combined to improve the reliability of pseudo-labels. Furthermore, the classification and alignment process is synergistically optimized through a fusion loss function, which will improve the fault identification accuracy and generalization ability of planetary gearboxes under complex and variable working conditions. Replacing traditional convolutional neural networks with convolutional spiking neural networks (CNNs) can more effectively acquire the impact and transient features of vibration signals, making it particularly suitable for identifying non-stationary signals such as gear meshing faults. A pulse feature continuity module is employed to transform pulse firing behavior into a stable, continuous feature representation, thereby effectively integrating CNNs with traditional statistical modeling and machine learning methods. A dual-branch structure with a pseudo-label branch and a distribution alignment branch is designed to decouple the target domain pseudo-label generation from the feature alignment process, avoiding the mutual interference between label noise and the alignment process in traditional methods, thus improving overall training stability and reliability. By setting a confidence screening strategy for target domain samples, pseudo-labels are dynamically updated during training, effectively reducing the risk of erroneous label propagation and improving the stability of model training and final recognition accuracy.

[0034] Example 2 like Figure 2 As shown, in Embodiment 2 of the present invention, a planetary gearbox fault diagnosis system is provided, the system comprising: Preprocessing module 1 is used to acquire the vibration acceleration signal of the planetary gearbox during operation, and to preprocess the vibration acceleration signal to obtain a pulse sequence dataset; Extraction module 2 is used to construct a convolutional spiking neural network feature extraction module, which extracts spatiotemporal features from the pulse sequence dataset to obtain pulse output; Pulse module 3 is used to determine the pulse firing rate of a single neuron based on the pulse output, and to determine the continuous feature representation based on the pulse firing rate; Branch module 4 is used to construct a dual-branch processing structure based on the continuous feature representation, construct a total loss function based on the dual-branch processing structure, and update the convolutional spiking neural network based on the total loss function and the backpropagation algorithm; The diagnostic module 5 is used to acquire the vibration acceleration signal of the target domain to be diagnosed, input the vibration acceleration signal of the target domain to be diagnosed into the updated convolutional pulse neural network for fault diagnosis, and output the fault diagnosis result.

[0035] Specifically, preprocessing module 1 is used for: The vibration acceleration signal is divided into several sub-samples using a sliding window. Each sub-sample is normalized, and the normalized sub-samples are mapped to pulse firing frequencies using firing rate encoding to obtain a pulse sequence dataset.

[0036] The convolutional spiking neural network feature extraction module adopts a serial cascaded and functionally layered design. The convolutional spiking neural network feature extraction module includes two convolutional modules, each of which is composed of a convolutional layer, a batch normalization layer, and a leak integral ignition layer connected in series. The dynamic changes in neuronal membrane potential and the pulse ignition process of the leakage integral ignition layer are described as follows: ; In the formula, for Membrane potential at time t, The membrane time constant is This is the resting potential. for Input current at any given moment; When membrane potential Exceeding the preset ignition threshold At this time, the neuron will generate a pulse signal, which will be transmitted through the step function. After encoding and outputting the pulses, the pulse outputs of the corresponding neurons in the corresponding layer of the convolutional spiking neural network are obtained: ; In the formula, For the first convolutional spiking neural network Layer One neuron in The pulse output at any given moment.

[0037] The pulse module 3 is used for: The pulse output is combined into multiple channels to obtain a multi-channel discrete pulse sequence, and a preset time window is determined. Calculate the spiking rate of a single neuron within a preset time window: ; In the formula, For the first convolutional spiking neural network Layer One neuron in Pulse output at any moment For the first convolutional spiking neural network Layer The firing rate of each neuron; The pulse firing rate is standardized, and the standardized pulse firing rates are combined in channel order to obtain a continuous feature vector. : ; In the formula, For the standardized convolutional spiking neural network, the first... Layer The firing rate of each neuron; The continuous feature vector Nonlinear mapping is performed in the input fully connected layer to obtain continuous feature representations. : ; In the formula, It is a non-linear activation function. These are learnable weights and learnable biases.

[0038] The dual-branch processing structure includes a pseudo-label branch and a feature distribution alignment branch. The pseudo-label generation branch is used to utilize unlabeled target domain data, input target domain features into a multi-layer fully connected network classifier, connect a Softmax function at the end, map the output to class probabilities in the 0-1 interval with a sum of 1, select samples with confidence scores higher than a set threshold as reliable samples based on the probabilities, and assign corresponding pseudo-labels. The target domain samples with pseudo-labels are then retrained to enhance the model's ability to discriminate target domain data. The feature distribution alignment branch is used to reduce the distribution difference between the source and target domains. It inputs source and target domain samples into the feature extraction network to obtain the corresponding feature representations. and and through the distribution metric function right , Constraints are applied to reduce the edge distribution differences between the source and target domains, including distribution alignment loss. for: .

[0039] Wherein, the total loss function for: ; In the formula, For pseudo-label loss, To balance the hyperparameters.

[0040] In other embodiments of the present invention, the present invention provides the following technical solution: a computer, including a memory 102, a processor 101, and a computer program stored in the memory 102 and executable on the processor 101, wherein the processor 101 executes the computer program to implement the planetary gearbox fault diagnosis method as described above.

[0041] Specifically, the processor 101 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or one or more integrated circuits that can be configured to implement the embodiments of the present invention.

[0042] The memory 102 may include a large-capacity memory for data or instructions. For example, and not limitingly, the memory 102 may include a hard disk drive (HDD), a floppy disk drive, a solid-state drive (SSD), flash memory, an optical disk drive, a magneto-optical disk drive, magnetic tape, or a Universal Serial Bus (USB) drive, or a combination of two or more of these. Where appropriate, the memory 102 may include removable or non-removable (or fixed) media. Where appropriate, the memory 102 may be internal or external to a data processing device. In a particular embodiment, the memory 102 is non-volatile memory. In a particular embodiment, the memory 102 includes read-only memory (ROM) and random access memory (RAM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable read-only memory (PROM), an erasable read-only memory (EPROM), an electrically erasable read-only memory (EEPROM), an electrically alterable read-only memory (EAROM), or flash memory, or a combination of two or more of these. Where appropriate, the RAM can be Static Random-Access Memory (SRAM) or Dynamic Random-Access Memory (DRAM). DRAM can be Fast Page Mode Dynamic Random Access Memory (FPMDRAM), Extended Data Out Dynamic Random Access Memory (EDODRAM), Synchronous Dynamic Random-Access Memory (SDRAM), etc.

[0043] The memory 102 can be used to store or cache various data files that need to be processed and / or used for communication, as well as possible computer program instructions executed by the processor 101.

[0044] The processor 101 implements the above-described planetary gearbox fault diagnosis method by reading and executing computer program instructions stored in the memory 102.

[0045] In some embodiments, the computer may further include a communication interface 103 and a bus 100. For example, Figure 3 As shown, the processor 101, memory 102, and communication interface 103 are connected through bus 100 and complete communication with each other.

[0046] The communication interface 103 is used to enable communication between the various modules, devices, units, and / or equipment in the embodiments of the present invention. The communication interface 103 can also enable data communication with other components such as external devices, image / data acquisition devices, databases, external storage, and image / data processing workstations.

[0047] Bus 100 includes hardware, software, or both, that couples components of a computer device together. Bus 100 includes, but is not limited to, at least one of the following: data bus, address bus, control bus, expansion bus, and local bus. For example, and not as a limitation, bus 100 may include an Accelerated Graphics Port (AGP) or other graphics bus, an Extended Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Hyper Transport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an InfiniBand interconnect, a Low Pin Count (LPC) bus, a memory bus, a Micro Channel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Bus (VLB) bus, or other suitable buses, or a combination of two or more of these. Where appropriate, bus 100 may include one or more buses. Although specific buses are described and illustrated in the embodiments of the present invention, the present invention is contemplated by any suitable bus or interconnect.

[0048] The computer can execute the planetary gearbox fault diagnosis method of the present invention based on the acquired planetary gearbox fault diagnosis system, thereby realizing planetary gearbox fault diagnosis.

[0049] In some further embodiments of the present invention, in conjunction with the above-described planetary gearbox fault diagnosis method, the present invention provides the following technical solution: a storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the above-described planetary gearbox fault diagnosis method.

[0050] Those skilled in the art will understand that the logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as a ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can mean any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0051] More specific examples of readable media (a non-exhaustive list) include: electrical connections (electronic devices) with one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0052] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0053] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0054] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention. Therefore, the protection scope of this invention patent should be determined by the appended claims.

Claims

1. A method for diagnosing faults in a planetary gearbox, characterized in that, include: The vibration acceleration signal of the planetary gearbox during operation is acquired, and the vibration acceleration signal is preprocessed to obtain a pulse sequence dataset; A convolutional spiking neural network feature extraction module is constructed to extract spatiotemporal features from the pulse sequence dataset to obtain pulse output. The pulse firing rate of a single neuron is determined based on the pulse output, and the continuous feature representation is determined based on the pulse firing rate. A dual-branch processing structure is constructed based on the continuous feature representation, a total loss function is constructed based on the dual-branch processing structure, and the convolutional spiking neural network is updated based on the total loss function and the backpropagation algorithm. The vibration acceleration signal of the target domain to be diagnosed is obtained, and the vibration acceleration signal of the target domain to be diagnosed is input into the updated convolutional pulse neural network for fault diagnosis, so as to output the fault diagnosis result.

2. The planetary gearbox fault diagnosis method according to claim 1, characterized in that, The step of preprocessing the vibration acceleration signal to obtain a pulse sequence dataset includes: The vibration acceleration signal is divided into several sub-samples using a sliding window. Each sub-sample is normalized, and the normalized sub-samples are mapped to pulse firing frequencies using firing rate encoding to obtain a pulse sequence dataset.

3. The planetary gearbox fault diagnosis method according to claim 1, characterized in that, The convolutional spiking neural network feature extraction module adopts a serial cascaded and functional layered design. The convolutional spiking neural network feature extraction module includes two convolutional modules. Each convolutional module is composed of a convolutional layer, a batch normalization layer, and a leaky integral ignition layer connected in series. The dynamic changes in neuronal membrane potential and the pulse ignition process of the leakage integral ignition layer are described as follows: ； In the formula, for Membrane potential at time t, The membrane time constant is This is the resting potential. for Input current at any given moment; When membrane potential Exceeding the preset ignition threshold At this time, the neuron will generate a pulse signal, which will be transmitted through the step function. After encoding and outputting the pulses, the pulse outputs of the corresponding neurons in the corresponding layer of the convolutional spiking neural network are obtained: ； In the formula, For the first convolutional spiking neural network Layer One neuron in The pulse output at any given moment.

4. The planetary gearbox fault diagnosis method according to claim 1, characterized in that, The steps of determining the pulse firing rate of a single neuron based on the pulse output and determining the continuous features based on the pulse firing rate include: The pulse output is combined into multiple channels to obtain a multi-channel discrete pulse sequence, and a preset time window is determined. Calculate the spiking rate of a single neuron within a preset time window: ； In the formula, For the first convolutional spiking neural network Layer One neuron in Pulse output at any moment For the first convolutional spiking neural network Layer The firing rate of each neuron; The pulse firing rate is standardized, and the standardized pulse firing rates are combined in channel order to obtain a continuous feature vector. : ； In the formula, For the standardized convolutional spiking neural network, the first... Layer The firing rate of each neuron; The continuous feature vector Nonlinear mapping is performed in the input fully connected layer to obtain continuous feature representations. : ； In the formula, It is a non-linear activation function. These are learnable weights and learnable biases.

5. The planetary gearbox fault diagnosis method according to claim 1, characterized in that, The dual-branch processing structure includes a pseudo-label branch and a feature distribution alignment branch; The pseudo-label generation branch is used to utilize unlabeled target domain data, input target domain features into a multi-layer fully connected network classifier, connect a Softmax function at the end, map the output to class probabilities in the 0-1 interval with a sum of 1, select samples with confidence scores higher than a set threshold as reliable samples based on the probabilities, and assign corresponding pseudo-labels. The target domain samples with pseudo-labels are then retrained to enhance the model's ability to discriminate target domain data. The feature distribution alignment branch is used to reduce the distribution difference between the source and target domains. It inputs source and target domain samples into the feature extraction network to obtain the corresponding feature representations. and and through the distribution metric function right , Constraints are applied to reduce the edge distribution differences between the source and target domains, including distribution alignment loss. for: 。 6. The planetary gearbox fault diagnosis method according to claim 5, characterized in that, The total loss function for: ； In the formula, For pseudo-label loss, To balance the hyperparameters.

7. A planetary gearbox fault diagnosis system, characterized in that, The system includes: The preprocessing module is used to acquire the vibration acceleration signal of the planetary gearbox during operation, and to preprocess the vibration acceleration signal to obtain a pulse sequence dataset. An extraction module is used to construct a convolutional spiking neural network feature extraction module, which extracts spatiotemporal features from the pulse sequence dataset to obtain pulse output. The pulse module is used to determine the pulse firing rate of a single neuron based on the pulse output, and to determine the continuous feature representation based on the pulse firing rate; The branching module is used to construct a dual-branch processing structure based on the continuous feature representation, construct a total loss function based on the dual-branch processing structure, and update the convolutional spiking neural network based on the total loss function and the backpropagation algorithm. The diagnostic module is used to acquire the vibration acceleration signal of the target domain to be diagnosed, input the vibration acceleration signal of the target domain to be diagnosed into the updated convolutional pulse neural network for fault diagnosis, and output the fault diagnosis result.

8. The planetary gearbox fault diagnosis system according to claim 7, characterized in that, The preprocessing module is used for: The vibration acceleration signal is divided into several sub-samples using a sliding window. Each sub-sample is normalized, and the normalized sub-samples are mapped to pulse firing frequencies using firing rate encoding to obtain a pulse sequence dataset.

9. A computer comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the planetary gearbox fault diagnosis method as described in any one of claims 1 to 6.

10. A storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the planetary gearbox fault diagnosis method as described in any one of claims 1 to 6.

Images