Hydraulic system fault diagnosis method based on improved MLP model and related device

Abstract

The application discloses a hydraulic system fault diagnosis method based on an improved MLP model and related devices, and relates to the technical field of fault diagnosis. The method obtains working condition monitoring data such as pressure, actuator speed and oil line flow of a hydraulic system, inputs the data into a trained improved MLP model and obtains a diagnosis result. In the fault diagnosis model, data enhancement preprocessing improves model robustness and relieves data imbalance through enhancement and standardization processing; a dynamic feature weighting module adaptively adjusts feature weights and enhances the contribution of key features; a feature cross module constructs high-order features through nonlinear transformation and residual connection and retains original information; and a classifier improves stability and prevents overfitting through batch normalization, GELU activation and Dropout mechanism. The method solves the problem that a traditional method relies on artificial feature engineering and has poor generalization ability, realizes efficient and accurate diagnosis of hydraulic system faults, and balances model lightweight and diagnosis accuracy.

Description

Technical Field

[0001] This application relates to the field of fault diagnosis technology, and in particular to a fault diagnosis method and related apparatus for hydraulic systems based on an improved MLP model. Background Technology

[0002] Fault modes in hydraulic systems exhibit highly nonlinear and multivariate coupling characteristics. In this context, selecting a suitable classification model to effectively capture the complex relationships between features is one of the key problems this research aims to address. Traditional machine learning methods, such as decision trees, support vector machines, and random forests, have shown good performance in certain pattern recognition tasks. Moreover, due to their relatively simple structure and low computational cost, these models have been widely used in industrial diagnostics. However, these methods typically rely on manual feature engineering and cannot automatically learn high-order nonlinear relationships between features, which limits their application in complex fault mode recognition. Furthermore, while decision trees and random forests can improve robustness through ensemble learning, model complexity increases rapidly with the amount of data. Deploying large ensemble models in embedded devices may be limited by computational resources.

[0003] In contrast, deep learning methods, especially classification models based on neural networks, demonstrate superior performance in handling complex data patterns due to their powerful automatic feature learning capabilities. Neural networks can extract features layer by layer through a multi-layered structure, enabling the model to learn implicit patterns in the data without relying on manually designed feature engineering. This end-to-end learning approach makes it more advantageous when dealing with high-dimensional and non-linear data. Specifically, the multilayer perceptron (MLP), as a fundamental and widely used neural network model, is the preferred choice for this study due to its strong feature modeling capabilities and low computational complexity. Through layer-by-layer computation and non-linear transformations of multiple neurons, the MLP can effectively uncover higher-order relationships between features, and enhance the model's generalization ability through reasonable parameter optimization and regularization strategies.

[0004] However, standard multilayer perceptrons still have some inherent limitations, such as the inability to dynamically adjust feature importance, a lack of ability to model feature cross-information, and a tendency to overfit. Therefore, proposing a lightweight, adaptively enhanced neural network model to better meet the fault diagnosis needs of hydraulic systems is essential for this field. Summary of the Invention

[0005] The purpose of this application is to provide a hydraulic system fault diagnosis method and related device based on an improved MLP model, which can achieve efficient and accurate diagnosis of hydraulic system faults, while taking into account both model lightweighting and fault diagnosis accuracy.

[0006] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a hydraulic system fault diagnosis method based on an improved MLP model, including: Acquire operational monitoring data of the hydraulic system to be identified; operational monitoring data includes key parameters such as hydraulic system pressure, actuator speed, and oil flow rate.

[0007] The operating condition monitoring data is input into the trained hydraulic system fault diagnosis model to obtain the hydraulic system fault diagnosis results. The hydraulic system fault diagnosis model is an improved MLP model based on dynamic feature weighting and feature cross-enhancement mechanisms. The hydraulic system fault diagnosis model includes a data augmentation preprocessing module, a dynamic feature weighting module, a feature cross-enhancement module, and a classifier. The data augmentation preprocessing module is used to enhance and standardize the original operating condition monitoring data to obtain preprocessed operating condition monitoring data. The dynamic feature weighting module is used to adaptively adjust the weights of each dimension of the preprocessed operating condition monitoring data. The feature cross-enhancement module is used to perform nonlinear transformation and cross-combination on the weights of the adaptively weighted features, and construct higher-order features through residual connections. The classifier is used to map the higher-order features to class probabilities and determine the hydraulic system fault diagnosis results through activation functions. The classifier is embedded with batch normalization, GELU activation, and Dropout mechanisms.

[0008] Optionally, before inputting the operating condition monitoring data into the trained hydraulic system fault diagnosis model to obtain the hydraulic system fault diagnosis result, the process also includes training the hydraulic system fault diagnosis model, including the following steps: Fault simulation is performed using a hydraulic system simulation model to obtain a hydraulic system fault data sample set. This set includes several hydraulic system fault data samples. Each sample contains operating condition monitoring data obtained from the fault simulation and the corresponding hydraulic system fault category. The fault categories are: normal operation, hydraulic pump failure, directional valve failure, throttle valve failure, or hydraulic cylinder failure.

[0009] The hydraulic system fault data sample set is input into the data augmentation preprocessing module. After data augmentation preprocessing, it is divided into training set, validation set and test set according to a preset ratio.

[0010] The model parameters of the hydraulic system fault diagnosis model are updated using backpropagation with the training set, and the performance of the trained hydraulic system fault diagnosis model is tested with the test set to perform hyperparameter tuning.

[0011] After training the hydraulic system fault diagnosis model on the training and test sets, the performance of the trained hydraulic system fault diagnosis model on new data is evaluated using the validation set.

[0012] Optionally, fault simulation is performed using a hydraulic system simulation model to obtain a hydraulic system fault data sample set, specifically including: Hydraulic system simulation models can be built using Simulink or Simscape Fluids.

[0013] The hydraulic system simulation model was used to investigate hydraulic pump failure, directional valve failure, throttle valve failure, or hydraulic cylinder failure, and the operating condition monitoring data of the hydraulic system simulation model was obtained. The operating condition monitoring data included hoisting cylinder speed, tilting cylinder speed, system pressure, hoisting cylinder pressure, tilting cylinder rodless chamber pressure, tilting cylinder rod chamber pressure, oil outlet flow rate, oil tank return flow rate, hoisting cylinder flow rate, tilting cylinder rodless chamber flow rate, and tilting cylinder rod chamber flow rate.

[0014] The operating condition monitoring data and the corresponding hydraulic system fault categories are combined to form a hydraulic system fault data sample, thus obtaining a hydraulic system fault data sample set.

[0015] Optionally, hydraulic pump failure, directional valve failure, throttle valve failure, or hydraulic cylinder failure can be simulated using a hydraulic system simulation model. Specifically, the simulation can be performed by adjusting the volumetric efficiency parameter of the hydraulic pump in the hydraulic system simulation model; the simulation can be performed by adjusting the leakage area parameter of the directional valve in the hydraulic system simulation model; the simulation can be performed by reducing the flow discharge coefficient in the hydraulic system simulation model; and the simulation can be performed by adding an external orifice in the hydraulic system simulation model.

[0016] Optionally, the dynamic feature weighting module adaptively adjusts the weights of each dimension of the preprocessed operating condition monitoring data as shown in the following equation: in, These are the feature vectors for each dimension of the preprocessed operating condition monitoring data. and These are the weight matrices of the two fully connected layers in the dynamic feature weighting module. and It is the corresponding bias term. It is a non-linear activation function; This represents the Sigmoid function, used to normalize weight values. This represents element-wise multiplication, which allows features to be dynamically adjusted according to the weights of different samples.

[0017] Optionally, the feature cross module performs nonlinear transformation and cross combination on the features of each dimension after adaptive weight adjustment, as shown in the following equation: in, These are the features of each dimension after adaptive weight adjustment; and These are the weight matrix and bias term of the first fully connected layer of the feature cross module, which are responsible for mapping the input features to the 12-dimensional latent space. and These are the weight matrix and bias term of the second fully connected layer of the feature cross module, which are responsible for converting the 12-dimensional latent features back to the original dimensions.

[0018] Secondly, this application provides a hydraulic system fault diagnosis system based on an improved MLP model, comprising: The operating condition monitoring data acquisition unit is used to acquire operating condition monitoring data of the hydraulic system to be identified; the operating condition monitoring data includes key parameters such as hydraulic system pressure, actuator speed and oil flow.

[0019] The hydraulic system fault diagnosis unit is used to input operating condition monitoring data into a trained hydraulic system fault diagnosis model to obtain hydraulic system fault diagnosis results. The hydraulic system fault diagnosis model is an improved MLP model based on dynamic feature weighting and feature cross-enhancement mechanisms. The hydraulic system fault diagnosis model includes a data augmentation preprocessing module, a dynamic feature weighting module, a feature cross-enhancement module, and a classifier. The data augmentation preprocessing module is used to enhance and standardize the original operating condition monitoring data to obtain preprocessed operating condition monitoring data. The dynamic feature weighting module is used to adaptively adjust the weights of each dimension of the preprocessed operating condition monitoring data. The feature cross-enhancement module is used to perform nonlinear transformation and cross-combination on the weights of the adaptively weighted features, constructing higher-order features through residual connections. The classifier is used to map higher-order features to class probabilities and determine the hydraulic system fault diagnosis results through activation functions. The classifier incorporates batch normalization, GELU activation, and Dropout mechanisms.

[0020] Thirdly, this application provides a computer device, 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 hydraulic system fault diagnosis method based on the improved MLP model described in any of the above claims.

[0021] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the hydraulic system fault diagnosis method based on the improved MLP model described in any of the preceding claims.

[0022] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the hydraulic system fault diagnosis method based on the improved MLP model described in any of the preceding claims.

[0023] According to the specific embodiments provided in this application, the following technical effects are disclosed: This application provides a hydraulic system fault diagnosis method and related apparatus based on an improved MLP model. The method acquires key parameters of the hydraulic system to be identified, such as pressure, actuator speed, and oil flow rate, to obtain operating condition monitoring data, providing comprehensive and accurate raw data support for subsequent fault diagnosis. The operating condition monitoring data is then input into a trained hydraulic system fault diagnosis model to obtain the fault diagnosis results. In this model, a data augmentation preprocessing module enhances and standardizes the original operating condition monitoring data, resulting in preprocessed data. This preprocessed data improves the model's robustness to different operating conditions, alleviates the problem of uneven data distribution, and ensures consistent feature dimensions, preventing some features from dominating model training due to excessively large dimensions, thus ensuring a relatively balanced influence of each feature on the model. Finally, a dynamic feature weighting module adaptively weights the features of each dimension of the preprocessed operating condition monitoring data. The adjustment method dynamically enhances the contribution of important features and suppresses irrelevant information, solving the problem of weakening key features caused by assigning fixed weights to all features in traditional MLP models, thus improving the model's ability to capture fault-sensitive features. Through a feature cross-module, based on adaptive weight adjustment, nonlinear transformations and cross-combinations are performed on the features of each dimension after weight adaptive adjustment. Higher-order features are constructed through residual connections, which can learn deep-level interaction relationships between features (such as the nonlinear correlation between pressure and flow) while preserving original feature information, avoiding information loss due to excessive transformation, and enhancing the model's feature representation ability. Finally, a classifier maps higher-order features to class probabilities, and an activation function determines the fault diagnosis results of the hydraulic system. Mechanisms such as batch normalization and Dropout integrated into the classifier stabilize data distribution, prevent overfitting, improve the model's generalization ability and classification accuracy, and ultimately achieve accurate diagnosis of hydraulic system faults. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0025] Figure 1 This is a flowchart illustrating a hydraulic system fault diagnosis method based on an improved MLP model, provided as an embodiment of this application.

[0026] Figure 2 This is a schematic diagram of the structure of the improved MLP model in a hydraulic system fault diagnosis method based on an improved MLP model provided in an embodiment of this application.

[0027] Figure 3 This is a schematic diagram of the dynamic feature weighting module in a hydraulic system fault diagnosis method based on an improved MLP model, provided in an embodiment of this application.

[0028] Figure 4 This is a schematic diagram of the structure of a feature cross module in a hydraulic system fault diagnosis method based on an improved MLP model, provided in an embodiment of this application.

[0029] Figure 5 This is a schematic diagram of the classifier structure in a hydraulic system fault diagnosis method based on an improved MLP model, provided in an embodiment of this application.

[0030] Figure 6 This is a schematic diagram illustrating the process of training a hydraulic system fault diagnosis model in a hydraulic system fault diagnosis method based on an improved MLP model, as provided in an embodiment of this application.

[0031] Figure 7 This is a schematic diagram of a simulation model of a forklift hydraulic system built in a hydraulic system fault diagnosis method based on an improved MLP model, provided in an embodiment of this application.

[0032] Figure 8 This is a comparison curve of the training performance of different models in a fault classification task in a hydraulic system fault diagnosis method based on an improved MLP model provided in an embodiment of this application.

[0033] Figure 9 This is a schematic diagram of the functional modules of a hydraulic system fault diagnosis system based on an improved MLP model, provided as an embodiment of this application.

[0034] Figure 10 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0035] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0036] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0037] This application provides a hydraulic system fault diagnosis method based on an improved MLP model. In one exemplary embodiment, such as... Figure 1 As shown, it includes the following steps: A1. Obtain the operating condition monitoring data of the hydraulic system to be identified; the operating condition monitoring data includes key parameters such as hydraulic system pressure, actuator speed, and oil flow.

[0038] A2. Input the operating condition monitoring data into the trained hydraulic system fault diagnosis model to obtain the hydraulic system fault diagnosis results. The hydraulic system fault diagnosis model is an improved MLP model based on dynamic feature weighting and feature cross-enhancement mechanisms. The hydraulic system fault diagnosis model includes a data augmentation preprocessing module, a dynamic feature weighting module, a feature cross-enhancement module, and a classifier. The data augmentation preprocessing module is used to enhance and standardize the original operating condition monitoring data to obtain preprocessed operating condition monitoring data. The dynamic feature weighting module is used to adaptively adjust the weights of each dimension of the preprocessed operating condition monitoring data. The feature cross-enhancement module is used to perform nonlinear transformation and cross-combination on the weights of the adaptively adjusted features, and construct higher-order features through residual connections. The classifier is used to map the higher-order features to class probabilities and determine the hydraulic system fault diagnosis results through activation functions. The classifier also incorporates batch normalization, GELU activation, and Dropout mechanisms.

[0039] To achieve high-accuracy forklift fault diagnosis while maintaining a lightweight model design, a lightweight neural network model, Enhanced-Lite-MLP, is proposed. This model combines dynamic feature weighting and feature cross-enhancement mechanisms to build a hydraulic system fault diagnosis model, such as... Figure 2As shown, the hydraulic system fault diagnosis model first enhances the original input data by incorporating Gaussian noise, random occlusion, and feature scaling to improve the model's robustness and alleviate the problem of imbalanced data distribution. The processed sample features are then adaptively adjusted by a dynamic feature weighting module. This module employs a two-layer multilayer perceptron structure, generating feature weights through Sigmoid activation to dynamically weight the input features, enhancing the contribution of important features and suppressing irrelevant information. The weighted features then enter a feature cross-validation module, which transforms the feature representation through two layers of MLP, learning deep-level interactions between features and preserving original feature information through residual connections, thereby improving the model's feature representation capability. The fused features are then input into a classifier, which includes a BatchNorm normalization layer, a GELU activation function, and a Dropout layer to improve the model's stability and generalization ability. Finally, a fully connected layer and a Softmax function output the fault class probability, achieving accurate classification of forklift hydraulic system faults.

[0040] Specifically, the model first enhances and standardizes the original working condition monitoring data through a data augmentation preprocessing module, and combines methods such as Gaussian noise, random occlusion, and feature scaling to improve the robustness of the model and alleviate the problem of uneven data distribution.

[0041] Data augmentation can improve the robustness of a model under different operating conditions and avoid performance degradation due to insufficient data or distribution bias. This embodiment employs three data augmentation strategies: Gaussian noise, random occlusion, and feature scaling.

[0042] In the original data Add random noise with a mean of 0 and a variance of 0.02 to the top. This allows for the simulation of sensor measurement errors and random changes in the environment, enabling the model to better adapt to uncertainties in real-world applications; through M mask One or two features are randomly selected from the training data and assigned a value of 0 to simulate sensor malfunctions or signal loss, improving the model's adaptability to feature loss and enabling accurate judgments even with partial data loss. The feature value of each sample is multiplied by a random number between [0.9, 1.1]. This allows the model to adapt to changes in data scale under different devices and operating conditions, improving generalization ability and ensuring consistency between training and real-world applications. Data augmentation can be represented as: Because the features in the sample data have different dimensions, features with larger dimensions may dominate the model's training process, causing the model to ignore features with smaller dimensions. Therefore, the augmented data is standardized using the following formula: Standardization is the mean of all features. μ Adjusted to 0, standard deviation σ Adjust to 1 to ensure that the distribution of each feature conforms to the standard normal distribution. These are the standardized features. Standardization ensures that the weights of each feature have a relatively balanced impact on the model, preventing certain features from negatively affecting model training due to excessively large or small dimensions.

[0043] Next, as Figure 3 As shown, the dynamic feature weighting module employs a two-layer perceptron structure, combining activation functions for feature mapping and nonlinear transformation. First, a fully connected layer (projection layer) maps the input features to an 8-dimensional latent feature space. h 1. As shown in the following formula: Then, the GELU activation function is applied to... h 1. Perform a nonlinear transformation to improve the model's nonlinear expressive power, as shown in the following equation: The GELU function not only retains the advantages of ReLU but also allows negative features to pass through with a certain probability, thereby improving the model's stability when dealing with complex nonlinear data. At the same time, GELU's gradient smoothness helps avoid the vanishing gradient problem during training, enabling the model to learn the complex relationships between features more effectively.

[0044] The output dimension of the second fully connected layer (reconstruction layer) is 11. h 2. Mapping back to an 11-dimensional weight vector ensures that each feature receives an independent weight. This allows for assigning different weights to different features without changing the number of features, thereby enhancing the model's adaptability to failure modes. By using the Sigmoid activation function, the weights of each feature are compressed to the range [0,1], ensuring that the weighting coefficients of the features are neither too large nor too small, as shown in the following equation: In the formula, w It is an 11-dimensional feature weight vector, where each element represents the final weight of the corresponding feature, reflecting the importance of the feature.

[0045] Weight vector w Input features x scaled Dot product, implementing weighted operation: Since the output range of the Sigmoid activation function is [0,1], the weights... w The weights might be close to 0, which could lead to some features contributing very little or even disappearing from the model's predictions. To avoid information loss, a weight of 0.5 is added to the weights to ensure that each feature retains at least 50% of its information, thus preventing features from being completely suppressed. Meanwhile, the weights... w The dynamic weighting effect is maintained, ensuring that each feature still contributes to the weighted representation.

[0046] In summary, the process by which the dynamic feature weighting module adaptively adjusts the weights of each dimension of the preprocessed operational condition monitoring data is shown in the following equation: in, These are the feature vectors for each dimension of the preprocessed operating condition monitoring data. and These are the weight matrices of the two fully connected layers in the dynamic feature weighting module. and It is the corresponding bias term. It is a non-linear activation function; This represents the Sigmoid function, used to normalize weight values. This represents element-wise multiplication, which allows features to be dynamically adjusted according to the weights of different samples.

[0047] The core of the dynamic weighting module lies in learning the importance of each input feature across different samples. Specifically, during training, each time a new sample is input, the model calculates a new set of feature weights based on the performance of the sample's features. These weights are continuously optimized as training progresses, helping the model gradually learn the feature weighting method best suited for the task. Therefore, dynamism is reflected not only in the network structure design but also in the adaptive adjustment mechanism during training.

[0048] Next, as Figure 4 As shown, the feature cross-multiplication module adopts a two-layer perceptron structure, mapping the input features from the original dimension to a latent space, and then restoring them to the original dimension. The mathematical model can be described as follows: In the formula, Features are those that have undergone dynamic feature weighting; and These are the weight matrix and bias term of the first fully connected layer, respectively, which are responsible for mapping the input features to the 12-dimensional latent space; and This transforms the latent features back to their original dimensions. The entire module applies the GELU activation function twice, which not only preserves some negative information but also facilitates gradient propagation, helping to capture complex feature combinations.

[0049] Setting the hidden layer dimension to 12, a slight increase from the original feature dimension, is primarily for two reasons: First, this setup creates a lightweight bottleneck structure, reducing the number of model parameters while still extracting key interaction information. Second, appropriately increasing the hidden layer dimension helps capture finer coupling relationships between sensor data. For example, for hydraulic pump failures, the nonlinear relationship between flow rate (Q1) and system pressure (P1) may be more significant, while for hydraulic cylinder failures, other feature combinations may exhibit different cross-patterns. Through cross-pattern modules, the model can construct composite features sensitive to different failure types with lower computational cost.

[0050] Another consideration for adopting a two-layer structure is to avoid information loss that may result from single-layer transformations. Single-layer networks often struggle to capture deep feature interactions. Stacking multiple layers can enhance expressive power but is detrimental to model lightweighting and can easily lead to problems such as gradient vanishing. Therefore, this study uses the GELU activation function in each layer and maps the output back to the original feature dimension. The dynamically weighted features are combined with the output through residual connections and used as input to the subsequent classifier. Residual connection (RC) technology provides a more direct propagation path for gradients by directly adding the module input and output, effectively alleviating the gradient vanishing problem in deep networks. Simultaneously, this design ensures that after multiple nonlinear transformations, the original input information retains some effective features, preventing information loss due to excessive transformation. Without residual connections, the output of the feature crossing module is directly used. When the feature cross module fails to learn or removes some key feature information as input to the classifier, the original feature information will be lost, affecting the classification performance.

[0051] Finally, as Figure 5 As shown, the classifier employs a two-layer perceptron and embeds batch normalization, GELU activation, and dropout mechanisms in the hidden layers to enhance the model's non-linear expressiveness and prevent overfitting. The classifier outputs a 5-dimensional logits vector. During the training phase, the cross-entropy loss function is directly input; during the inference phase, the probability distribution of each category is obtained through softmax transformation, and the final classification result is determined using the argmax operation.

[0052] The classifier's input features are fused through residual connections, combining the results of feature weighting and feature cross-interaction, maintaining a dimension of 11. The first fully connected layer maps the input features to a 16-dimensional hidden layer, thus providing the model with stronger non-linear expressive power and enabling it to extract more valuable information from the input features.

[0053] Considering that linear transformations may lead to unstable output data distribution, with some dimensions potentially being too large or too small, affecting subsequent nonlinear transformations, batch normalization (BatchNorm) was applied to the features. In the formula, It is a very small number (e.g.) ), to prevent the denominator from being 0; x ij For the first in a batch of training data i The j-th eigenvalue of a eigenvector; μ j It is the mean of the j-th dimension feature of this batch of training data: ; It is the first batch of training data Variance obtained from dimensional features: Since the normalized data distribution may be too strict, BatchNorm allows learning a new set of parameters. and Adjustments were made to ensure the data still retains a degree of flexibility: In the formula, (scale) controls the scaling of features. (shift) controls the offset of the feature.

[0054] Final output y ij It still has a relatively stable distribution, but the model can automatically adjust based on the training data. , Enhanced expressive power. Batch Normalization (BN) performs batch normalization on each batch of data as it passes through a certain layer of the neural network, making its mean 0 and variance 1. A non-linear transformation using the GELU activation function ensures stable gradient propagation, thereby accelerating convergence and reducing the model's sensitivity to fluctuations in the distribution of different batches of data.

[0055] During the learning process of a neural network, the model may develop an overly strong memory of the training data due to an excessive number of parameters, leading to a decline in generalization ability—that is, the model performs well on the training set but poorly on the test set. To prevent the risk of overfitting on a limited number of samples, the classifier module employs the Dropout strategy in the hidden layers. During training, Dropout randomly discards 30% of the neuron outputs, essentially setting the outputs of some neurons to zero during forward propagation. This forces the network to learn more diverse feature combinations rather than relying excessively on certain neurons, thereby enhancing the model's robustness. Furthermore, Dropout has an effect similar to ensemble learning because the network structure changes slightly with each forward propagation, effectively training multiple different sub-models. This design makes the final model more stable during testing and avoids overfitting caused by over-reliance on certain neurons. Therefore, in fault classification tasks, Dropout ensures that the model learns more generalizable features, maintaining high classification accuracy even when faced with different types of fault data.

[0056] The classifier outputs a 5-dimensional logits vector through a second fully connected layer, with each element corresponding to a predicted score for a fault state. During model training, the output vector is directly input into the cross-entropy loss function. Internally, the cross-entropy loss function first performs a softmax transformation on the logits to calculate a normalized probability distribution, and then calculates the loss based on the true class label (a single numerical value). Although each sample corresponds to only one true label, because the output is a 5-dimensional probability vector, the loss function automatically selects the probability value corresponding to the correct class to calculate the loss, thus guiding the model parameters to update in the correct direction. During the inference phase, the output probability vector is subjected to a `torch.argmax()` operation to find the index corresponding to the maximum probability; the index value is the final classification result (class 0-4), thus determining the predicted fault class.

[0057] Before applying the trained hydraulic system fault diagnosis model in step A2 above, the hydraulic system fault diagnosis method should also include a process for training the hydraulic system fault diagnosis model. In a specific embodiment, when applying this fault diagnosis method to the fault diagnosis of a forklift's hydraulic system, obtaining real fault data is a major challenge in forklift hydraulic system fault diagnosis research. Due to the complex operating environment of forklift hydraulic systems, actual fault occurrences are random, and on-site fault data collection is costly and time-consuming. Therefore, to obtain sufficient training samples, a hydraulic system simulation model is established using SimscapeFluids, and faults are injected into key components such as hydraulic pumps, directional valves, throttle valves, and hydraulic cylinders. System response data under different operating conditions is obtained through simulation. Finally, the collected data is analyzed to extract effective fault features, providing support for subsequent fault diagnosis model training. Figure 6 As shown, the process includes the following steps: B1. Perform fault simulation using a hydraulic system simulation model to obtain a hydraulic system fault data sample set. The hydraulic system fault data sample set includes several hydraulic system fault data samples. The hydraulic system fault data samples include operating condition monitoring data obtained from the fault simulation and the corresponding hydraulic system fault categories. The hydraulic system fault categories are normal, hydraulic pump fault, directional valve fault, throttle valve fault, or hydraulic cylinder fault. Specifically, in this embodiment, step B1 includes: B11. Build a hydraulic system simulation model using Simulink or Simscape Fluids. The built hydraulic system simulation model is as follows: Figure 7 As shown, hydraulic pipelines are used to transmit the flow of hydraulic oil; control signals are used to control the spool position of the directional valve and the displacement adjustment of the hydraulic pump. This simulation model not only includes hydraulic valves that control flow, direction, and pressure, but also integrates multiple sensors to monitor the operating status of the hydraulic system in real time.

[0058] B12. Using a hydraulic system simulation model, simulate hydraulic pump failure, directional valve failure, throttle valve failure, or hydraulic cylinder failure, and obtain the operating condition monitoring data of the hydraulic system simulation model. The operating condition monitoring data includes hoisting cylinder speed, tilting cylinder speed, system pressure, hoisting cylinder pressure, tilting cylinder rodless chamber pressure, tilting cylinder rod chamber pressure, oil outlet flow rate, oil tank return flow rate, hoisting cylinder flow rate, tilting cylinder rodless chamber flow rate, and tilting cylinder rod chamber flow rate.

[0059] Specifically, in this embodiment, the hydraulic pump failure is simulated by adjusting the volumetric efficiency parameter of the hydraulic pump in the hydraulic system simulation model; the directional valve failure is simulated by adjusting the leakage area parameter of the directional valve in the hydraulic system simulation model; the throttle valve failure is simulated by reducing the flow discharge coefficient in the hydraulic system simulation model; and the hydraulic cylinder failure is simulated by adding an external orifice in the hydraulic system simulation model.

[0060] Specifically, hydraulic pump failure mainly manifests as internal leakage. Volumetric efficiency, which measures the ratio of the pump's output flow rate to its theoretical flow rate during actual operation, reflects the degree of internal leakage, i.e., its effective ability to deliver hydraulic oil. To simulate internal leakage, the volumetric efficiency parameter of the hydraulic pump was adjusted, setting three different levels of failure conditions. Under normal operating conditions, the volumetric efficiency is close to 100% (i.e., no leakage). For slight leakage, the volumetric efficiency drops to 85%, simulating a small-scale internal leak. For moderate leakage, the volumetric efficiency further decreases to 75%, simulating a more severe internal leak. For severe leakage, the volumetric efficiency drops to 65%, indicating a significant decrease in the hydraulic system's oil supply capacity.

[0061] Directional control valve malfunctions primarily manifest as internal leakage. By adjusting the leakage area parameter, different degrees of internal leakage can be simulated, with three distinct fault levels. Under normal operating conditions, the leakage area is set to 0, indicating virtually no leakage. In cases of slight leakage, the leakage area increases to 0.1 mm. 2 In moderate leakage conditions, the leakage area increases to 0.5 mm. 2 In cases of severe leakage, the leakage area increases to 1 mm. 2 Or larger.

[0062] Throttling valve malfunctions are mainly manifested as blockage of the throttle valve, resulting in a decrease in the flow coefficient. C v This value characterizes the resistance a fluid encounters when passing through a throttling element. A smaller value indicates greater flow resistance and lower flow rate, as shown in the following equation: in, A Let be the cross-sectional area of ​​the pipe. ΔP The pressure difference between the two ends of the pipe, and the flow rate. Q and C v They are directly proportional, therefore they can be reduced C v This is used to simulate the degree of blockage in a throttle valve.

[0063] In the throttle valve fault simulation, under normal circumstances, the flow discharge coefficient is set to the rated value of 0.7; when the throttle valve experiences slight blockage, Cv Reduced to 0.5; under moderate congestion conditions, Cv It further decreased to 0.3; while in cases of severe congestion, Cv The value was reduced to 0.1 to simulate the extreme case of limited traffic.

[0064] Hydraulic cylinder failure mainly manifests as external leakage. This is simulated by adding an external orifice to the simulation model, and three different failure levels are set, starting with minor leakage (1 mm). 2 Medium leakage 2 mm 2 3 mm of severe leakage 2 .

[0065] Under each fault condition, three fault levels were set, and the changing trends of parameters such as system pressure, flow rate, and hydraulic cylinder speed were observed and corresponding curves were plotted to analyze the impact of the fault on the hydraulic system performance. Throughout this process, all studied faulty components participated in the normal operation of the hydraulic system. Table 1 compares the key parameters of the simulation results with the theoretical calculation results through simulation of normal operating conditions. The errors between the simulation results and the theoretical calculation results for each parameter do not exceed 1%. The existence of these errors is due to the inclusion of minor disturbance factors in the components of the simulation system. This verifies the accuracy of the simulation results and the effectiveness of the model.

[0066] Table 1 Simulation results and theoretical results under normal operating conditions

[0067] B13. Combine the operating condition monitoring data and the corresponding hydraulic system fault categories to form a hydraulic system fault data sample, thus obtaining a hydraulic system fault data sample set. In this embodiment, 800 forklift hydraulic system operation samples were finally collected through fault simulation experiments, as shown in Table 2.

[0068] Table 2. Hydraulic system fault data sample set

[0069] The data format includes a first column for sample number (1-800) and a last column for fault labels (0-4). Label 0 indicates a normal sample, while labels 1-4 indicate hydraulic pump fault, directional valve fault, throttle valve fault, and hydraulic cylinder fault, respectively. The middle 11 fields are real-time monitoring data from hydraulic system sensors. Specific features include: hoisting cylinder speed (V1), tilting cylinder speed (V2), system pressure (P1), hoisting cylinder pressure (P2), tilting cylinder rodless chamber pressure (P3), tilting cylinder rod chamber pressure (P4), oil outlet flow rate (Q1), oil tank return flow rate (Q2), hoisting cylinder flow rate (Q3), tilting cylinder rodless chamber flow rate (Q4), and tilting cylinder rod chamber flow rate (Q5). The data acquisition covers key parameters such as system pressure, actuator speed, and oil flow rate, exhibiting typical characteristics of industrial process monitoring data.

[0070] B2. Input the hydraulic system fault data sample set into the data augmentation preprocessing module. After data augmentation preprocessing, divide it into training set, validation set and test set according to the preset ratio.

[0071] In the training process of machine learning and deep learning models, dataset partitioning is a crucial step. Datasets are typically divided into training, testing, and validation sets. The main purpose is to ensure the model performs well on unseen data, thereby improving its generalization ability. The training set is used for model learning, the testing set for hyperparameter tuning and model selection, and the validation set for final evaluation of the model's performance and generalization ability.

[0072] In this embodiment, the dataset is divided according to a common ratio: 60% of the data is used as the training set, 20% as the validation set, and the remaining 20% ​​as the test set. This division ensures that the model can effectively learn the patterns in the data during multiple training iterations, and the model's performance is evaluated using the test set, thereby adjusting the model's hyperparameters, such as the learning rate and batch size, to avoid overfitting or underfitting. The validation set is completely independent and is used to finally evaluate the model's performance on new data, ensuring that the model has not "seen" the test data during training and avoiding evaluation bias.

[0073] B3. The model parameters of the hydraulic system fault diagnosis model are updated using backpropagation with the training set, and the performance of the trained hydraulic system fault diagnosis model is tested with the test set to perform hyperparameter tuning.

[0074] The training process of a model is essentially a parameter optimization process. The goal is to continuously adjust the network weights so that the model can predict the state of the hydraulic system as accurately as possible. The training process can be divided into three steps: (1) Forward propagation: The input data passes through the neural network to obtain the prediction result.

[0075] During the forward propagation phase, the model ultimately outputs a 5-dimensional vector. p (Probability distribution), each value in the vector p i This represents the probability that a sample belongs to a certain category. Typically, samples are batched during training. This refers to a group of samples input into the model during each forward and backward propagation. Each batch consists of different samples, exposing the model to diverse data distributions and reducing the risk of overfitting. Furthermore, batching fully utilizes GPU (Graphics Processing Unit) parallel computing, improving training efficiency.

[0076] (2) Loss calculation: Calculate the error between the predicted value and the true label to obtain the loss value.

[0077] When calculating the loss, the mean of all samples within the batch is used to ensure that the weights are not drastically affected by outliers of a single sample during each update. This study uses the cross-entropy loss function to calculate the loss: in, For the true label (hot-one encoding), if the sample Category ,but ,otherwise ; The model predicts the number of... k The probability of a class; N In this embodiment, the batch size is set to 64 samples. To mitigate the impact of class imbalance on the model's final prediction performance, class weights are introduced into the cross-entropy loss function. w k : Where, max( N () represents the number of samples in the category with the most samples in the dataset. N k Let k be the number of samples in class k. After introducing class weights, the loss calculation formula is: in Only when the true category is 1, the above formula can be simplified to: in, c It is a sample iThe true class is determined by the cross-entropy loss, which measures the difference between the true class distribution and the model's predicted distribution. It is suitable for multi-class classification problems. The main purpose of using a logarithmic loss function is to impose a larger loss on incorrect predictions (low-probability outputs). For example, when the predicted probability of the correct class is high (e.g., 0.9), -log(0.9) ≈ 0.105; while when the predicted probability is low (e.g., 0.1), the loss increases to -log(0.1) ≈ 2.302. This design encourages the model to continuously improve the prediction confidence of the correct class during training. It is important to note that the logarithmic function defaults to using the natural logarithm as the base. ln , and the exponential function in softmax e z Maintaining consistency simplifies gradient calculation. The introduction of the negative sign ensures that the optimization objective is to minimize the loss, i.e., to maximize the prediction probability of the correct class, which aligns with optimization intuition.

[0078] (3) Backpropagation: The gradient of the loss with respect to the parameters of each layer is calculated by the chain rule to guide the update of the model parameters.

[0079] After forward propagation and loss calculation are completed, the model enters the backpropagation phase. The goal of this phase is to use the chain rule to backpropagate the gradient of the loss function with respect to the parameters of each layer, guiding the parameter updates and gradually reducing the error between the model output and the true label.

[0080] B4. After training the hydraulic system fault diagnosis model on the training and test sets, use the validation set to evaluate the performance of the trained hydraulic system fault diagnosis model on new data.

[0081] In this embodiment, the training performance of each model in the fault classification task is analyzed using four key metrics: training loss, training accuracy, validation loss, and validation accuracy.

[0082] Training loss measures the model's loss value on the training set; the lower the loss, the better the model fits the training data. Training accuracy measures the model's classification accuracy on the training set; the higher the accuracy, the stronger the model's ability to identify training samples. Validation loss is used to evaluate the model's generalization ability on the validation set; a lower loss value indicates that the model has good adaptability to unseen samples. Validation accuracy reflects the model's classification precision on the validation set; a higher accuracy indicates that the model can effectively identify different failure modes.

[0083] In the comparative experiment, the training performance of each model is as follows: Figure 8As shown, 1D-CNN represents a one-dimensional convolutional neural network model; RBFNN represents a radial basis function neural network; Attention-MLP represents a multilayer perceptron model with an added attention mechanism; and EnhancedLiteMLP is the improved MLP model proposed in this embodiment. Overall, all models exhibited rapid convergence in the initial training phase (the first 20 epochs), subsequently stabilizing. The improved MLP proposed in this application demonstrates superior performance at approximately 40 epochs and maintains its advantage throughout subsequent training. Due to the adoption of a dynamic learning rate scheduling strategy, the model's loss curve decreases rapidly in the initial training phase, exhibiting strong convergence. Although the loss value fluctuates somewhat between 40 and 60 epochs, the model training stabilizes as the learning rate gradually decays.

[0084] Based on the same inventive concept, this application also provides a system for implementing the hydraulic system fault diagnosis method based on the improved MLP model described above. The solution provided by this system is similar to the implementation described in the above method. In an exemplary embodiment, such as... Figure 9 As shown, a hydraulic system fault diagnosis system based on an improved MLP model is provided, including: The operating condition monitoring data acquisition unit is used to acquire operating condition monitoring data of the hydraulic system to be identified; the operating condition monitoring data includes key parameters such as hydraulic system pressure, actuator speed and oil flow.

[0085] The hydraulic system fault diagnosis unit is used to input operating condition monitoring data into a trained hydraulic system fault diagnosis model to obtain hydraulic system fault diagnosis results. The hydraulic system fault diagnosis model is an improved MLP model based on dynamic feature weighting and feature cross-enhancement mechanisms. The hydraulic system fault diagnosis model includes a data augmentation preprocessing module, a dynamic feature weighting module, a feature cross-enhancement module, and a classifier. The data augmentation preprocessing module is used to enhance and standardize the original operating condition monitoring data to obtain preprocessed operating condition monitoring data. The dynamic feature weighting module is used to adaptively adjust the weights of each dimension of the preprocessed operating condition monitoring data. The feature cross-enhancement module is used to perform nonlinear transformation and cross-combination on the weights of the adaptively weighted features, constructing higher-order features through residual connections. The classifier is used to map higher-order features to class probabilities and determine the hydraulic system fault diagnosis results through activation functions. The classifier incorporates batch normalization, GELU activation, and Dropout mechanisms.

[0086] certainly, Figure 9 The architecture shown is merely exemplary; it can be omitted as needed when implementing different functionalities. Figure 9 One or at least two components of the system shown.

[0087] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 10 As shown, the computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it can implement the hydraulic system fault diagnosis method based on an improved MLP model provided in the above embodiment.

[0088] Those skilled in the art will understand that Figure 10 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0089] In one exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0090] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0091] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0092] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0093] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0094] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0095] 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.

[0096] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A fault diagnosis method for hydraulic systems based on an improved MLP model, characterized in that, include: Acquire the operating condition monitoring data of the hydraulic system to be identified; The operating condition monitoring data includes hydraulic system pressure, actuator speed, and oil flow rate; The operating condition monitoring data is input into a trained hydraulic system fault diagnosis model to obtain hydraulic system fault diagnosis results. The hydraulic system fault diagnosis model is an improved MLP model based on dynamic feature weighting and feature cross-enhancement mechanisms. The hydraulic system fault diagnosis model includes a data augmentation preprocessing module, a dynamic feature weighting module, a feature cross-enhancement module, and a classifier. The data augmentation preprocessing module is used to enhance and standardize the original operating condition monitoring data to obtain preprocessed operating condition monitoring data. The dynamic feature weighting module is used to adaptively adjust the weights of each dimension of the preprocessed operating condition monitoring data. The feature cross-enhancement module is used to perform nonlinear transformation and cross-combination on the weights of the adaptively adjusted features, constructing higher-order features through residual connections. The classifier is used to map higher-order features to class probabilities and determine the hydraulic system fault diagnosis results through activation functions. The classifier incorporates batch normalization, GELU activation, and Dropout mechanisms.

2. The hydraulic system fault diagnosis method based on the improved MLP model according to claim 1, characterized in that, Before inputting the operating condition monitoring data into the trained hydraulic system fault diagnosis model to obtain the hydraulic system fault diagnosis result, the process also includes training the hydraulic system fault diagnosis model, including the following steps: Fault simulation is performed using a hydraulic system simulation model to obtain a hydraulic system fault data sample set. The hydraulic system fault data sample set includes several hydraulic system fault data samples. The hydraulic system fault data samples include operating condition monitoring data obtained from fault simulation and corresponding hydraulic system fault categories. The hydraulic system fault categories are normal, hydraulic pump fault, reversing valve fault, throttle valve fault, or hydraulic cylinder fault. The hydraulic system fault data sample set is input into the data augmentation preprocessing module. After data augmentation preprocessing, it is divided into training set, validation set and test set according to a preset ratio. The model parameters of the hydraulic system fault diagnosis model are updated using backpropagation with the training set, and the performance of the trained hydraulic system fault diagnosis model is tested with the test set, and hyperparameter tuning is performed. After training the hydraulic system fault diagnosis model using the training and test sets, the performance of the trained hydraulic system fault diagnosis model on new data is evaluated using the validation set.

3. The hydraulic system fault diagnosis method based on the improved MLP model according to claim 2, characterized in that, Fault simulation is performed using a hydraulic system simulation model to obtain a hydraulic system fault data sample set, specifically including: Build a hydraulic system simulation model using Simulink or Simscape Fluids; The hydraulic system simulation model is used to simulate hydraulic pump failure, directional valve failure, throttle valve failure, or hydraulic cylinder failure, and to obtain the operating condition monitoring data of the hydraulic system simulation model. The operating condition monitoring data includes hoisting cylinder speed, tilting cylinder speed, system pressure, hoisting cylinder pressure, tilting cylinder rodless chamber pressure, tilting cylinder rod chamber pressure, oil outlet flow rate, oil tank return flow rate, hoisting cylinder flow rate, tilting cylinder rodless chamber flow rate, and tilting cylinder rod chamber flow rate. The operating condition monitoring data and the corresponding hydraulic system fault categories are combined to form a hydraulic system fault data sample, thus obtaining a hydraulic system fault data sample set.

4. The hydraulic system fault diagnosis method based on the improved MLP model according to claim 3, characterized in that, The hydraulic system simulation model is used to simulate hydraulic pump failures, directional valve failures, throttle valve failures, or hydraulic cylinder failures. Specifically, the simulation is performed by adjusting the volumetric efficiency parameter of the hydraulic pump in the simulation model; by adjusting the leakage area parameter of the directional valve in the simulation model; by reducing the flow discharge coefficient in the simulation model; and by adding an external orifice in the simulation model.

5. The hydraulic system fault diagnosis method based on the improved MLP model according to claim 1, characterized in that, The dynamic feature weighting module adaptively adjusts the weights of each dimension of the preprocessed operating condition monitoring data as shown in the following formula: in, These are the feature vectors for each dimension of the preprocessed operating condition monitoring data. and These are the weight matrices of the two fully connected layers in the dynamic feature weighting module. and It is the corresponding bias term. It is a non-linear activation function; This represents the Sigmoid function, used to normalize weight values. This represents element-wise multiplication, which allows features to be dynamically adjusted according to the weights of different samples.

6. The hydraulic system fault diagnosis method based on the improved MLP model according to claim 1, characterized in that, The process by which the feature cross module performs nonlinear transformation and cross combination on the features of each dimension after adaptive weight adjustment is shown in the following equation: in, These are the features of each dimension after adaptive weight adjustment; and These are the weight matrix and bias term of the first fully connected layer of the feature cross module, which are responsible for mapping the input features to the 12-dimensional latent space. and These are the weight matrix and bias term of the second fully connected layer of the feature cross-module, which are responsible for converting the 12-dimensional hidden features back to the original dimensions.

7. A hydraulic system fault diagnosis system based on an improved MLP model, characterized in that, include: The working condition monitoring data acquisition unit is used to acquire the working condition monitoring data of the hydraulic system to be identified; The operating condition monitoring data includes hydraulic system pressure, actuator speed, and oil flow rate; A hydraulic system fault diagnosis unit is used to input the operating condition monitoring data into a trained hydraulic system fault diagnosis model to obtain hydraulic system fault diagnosis results. The hydraulic system fault diagnosis model is an improved MLP model based on dynamic feature weighting and feature cross-enhancement mechanisms. The hydraulic system fault diagnosis model includes a data augmentation preprocessing module, a dynamic feature weighting module, a feature cross-enhancement module, and a classifier. The data augmentation preprocessing module is used to enhance and standardize the original operating condition monitoring data to obtain preprocessed operating condition monitoring data. The dynamic feature weighting module is used to adaptively adjust the weights of each dimension of the preprocessed operating condition monitoring data. The feature cross-enhancement module is used to perform nonlinear transformation and cross-combination on the weights of the adaptively adjusted features, constructing higher-order features through residual connections. The classifier is used to map the higher-order features to class probabilities and determine the hydraulic system fault diagnosis results through an activation function. The classifier incorporates batch normalization, GELU activation, and Dropout mechanisms.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the hydraulic system fault diagnosis method based on the improved MLP model according to any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the hydraulic system fault diagnosis method based on the improved MLP model as described in any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the hydraulic system fault diagnosis method based on the improved MLP model as described in any one of claims 1-6.

Images