Cross-working-condition fault diagnosis method and system for gearbox of mining shovel loading robot

Abstract

The invention provides a cross-working-condition fault diagnosis method and system for a gearbox of a mining shovel loading robot. The method comprises the steps that S1, vibration signals of the gearbox of a lifting mechanism of the mining shovel loading robot under different working conditions are collected; s2, constructing a diagnosis network and performing adaptive feature extraction on the vibration signals; the diagnosis network comprises a feature extractor and a fault classifier of the one-dimensional deformable convolutional network; s3, training the diagnosis network by adopting a dual-stage domain adaptive strategy fused with triple loss, and outputting a diagnosis model; and S4, performing fault diagnosis on the vibration signal under the target working condition according to the diagnosis model, and outputting a diagnosis result. According to the method, a double-stage domain self-adaptive strategy is adopted, feature optimization is carried out on the source domain in the first stage, data enhancement and domain invariant feature learning are carried out through mixed source domain and target domain labeling samples in the second stage, the domain offset problem caused by working condition changes is effectively relieved, and cross-working-condition efficient fault diagnosis is achieved.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis, specifically to a method and system for cross-working-condition fault diagnosis of gearboxes in mining shovel-loading robots, and more particularly to a method and system for cross-working-condition fault diagnosis of gearboxes in the hoisting mechanism of mining shovel-loading robots. Background Technology

[0002] In mining operations, shovel-loading robots are key equipment in large-scale open-pit mining. Their lifting mechanisms undertake the tasks of digging and loading heavy materials, resulting in high workload and complex operating conditions, which places extremely high demands on the stability of the transmission system. The gearbox in the lifting mechanism of the electric shovel robot, as the core component for power transmission, is subjected to high loads, frequent starts and stops, and harsh environments such as dust and humidity for extended periods. This makes it highly susceptible to failures such as gear wear, pitting, tooth breakage, and bearing damage, leading to decreased transmission efficiency, increased vibration, and in severe cases, equipment shutdown or even safety accidents.

[0003] The gearboxes of the hoisting mechanisms in mining loader robots often operate under complex and variable conditions, including varying speeds, loads, speed increases and decreases, and high noise levels. In such environments, traditional fault diagnosis methods, such as vibration analysis and acoustic testing, are easily affected by environmental noise, limiting their diagnostic accuracy. Therefore, achieving high-precision cross-condition diagnosis of gearbox faults in the hoisting mechanisms of mining loader robots has become a key challenge for ensuring safe equipment operation and promoting the development of intelligent operation and maintenance in mining.

[0004] Patent document CN115409110A discloses a cross-condition gearbox fault diagnosis method, including: acquiring a large amount of labeled source domain fault data and some unlabeled target domain fault data; performing wavelet packet transform on the collected data to obtain wavelet packet coefficients, and using the wavelet packet coefficients as model input; constructing a diagnostic model, including a feature extractor, a domain discriminator, and a classifier, using the feature extractor to extract features from the source domain fault data and the target domain fault data respectively; performing weighted processing on the source domain fault data and then conducting adversarial training with the target domain fault data through the domain discriminator, calculating the adversarial loss, and calculating the classification loss through the classification results of the classifier, optimizing the parameters of the diagnostic model through the adversarial loss and the classification loss; performing wavelet packet transform on the data to be detected, inputting the wavelet coefficients into the model, and the model outputting the fault diagnosis result. The core of this patent document relies on wavelet packet transform to extract fixed frequency band features and achieves domain alignment through adversarial training. However, it suffers from two key limitations: first, wavelet packet transform feature extraction depends on manually selecting the number of decomposition layers and frequency band divisions, making it difficult to adapt to the non-stationarity and frequency shifts of vibration signals under complex mining conditions; second, adversarial training easily leads to pattern collapse or training instability, and cannot guarantee the intra-class compactness and inter-class separability of the feature space, affecting the model's interpretability and diagnostic robustness. In contrast, this invention employs a one-dimensional deformable convolutional network, which can adaptively and dynamically adjust the sampling position and weights of the convolutional kernel according to the local waveform features of the input signal, achieving adaptive feature extraction of vibration signals. This better suits the non-stationar characteristics of signals from mining equipment under conditions such as impact loads and frequent start-stop cycles. Furthermore, this invention actively constructs a compact intra-class structure and clear inter-class boundaries in the first stage through triplet loss, avoiding the feature chaos that may occur in adversarial training and providing a structured feature space foundation for subsequent active learning.

[0005] Patent document CN119719862A discloses a method and system for cross-condition fault diagnosis of planetary gearboxes driven by analog-digital fusion. The method includes: acquiring vibration, speed, and torque signals from the planetary gearbox; preprocessing the vibration signals; monitoring the preprocessed data to determine if the operating conditions have changed; updating the corresponding parameters of the dynamic model based on the acquired speed and torque signals when the operating conditions change, and generating simulation data matching the current operating conditions; inputting the simulation data into a first model (divided into analog signals, source domain signals, and target domain signals) to train the parameters in the first model; and acquiring real-time speed and torque signals from the planetary gearbox and inputting them into the trained first model for fault diagnosis. The core of this patent document relies on auxiliary signals such as speed and torque to construct a dynamic model and generates simulation data matching the current operating conditions through simulation. However, it has engineering limitations: firstly, it requires additional acquisition of speed and torque signals, increasing sensor deployment costs and system complexity; secondly, it relies on the accuracy of the dynamic model, and when the actual operating conditions deviate from the model assumptions, the representativeness of the simulation data decreases, leading to a decline in diagnostic performance. In contrast, this invention relies solely on vibration signals for fault diagnosis, requiring no auxiliary signals or physical models. This makes it suitable for the practical situation where mining loader robots face difficulties in deploying multi-source sensors in mining environments. Furthermore, this invention selects the most representative samples in the target domain for annotation through an active learning mechanism, combining data augmentation and triplet loss to achieve domain adaptation. This avoids the error propagation problem associated with simulation data generation, making it more suitable for engineering scenarios where annotation costs are high and samples are scarce. Summary of the Invention

[0006] To address the shortcomings of existing technologies, the purpose of this invention is to provide a method and system for cross-condition fault diagnosis of gearboxes in mining shovel loading robots.

[0007] A method for cross-condition fault diagnosis of gearbox in a mining shovel robot according to the present invention includes: Step S1: Collect vibration signals of the gearbox of the hoisting mechanism of the mining shovel robot under different working conditions; Step S2: Construct a diagnostic network and perform adaptive feature extraction on the vibration signal; The diagnostic network includes a feature extractor and a fault classifier from a one-dimensional deformable convolutional network. Step S3: Based on the extracted features, the diagnostic network is trained using a two-stage domain adaptive strategy that incorporates triplet loss, and the diagnostic model is output. Step S4: Perform fault diagnosis on the vibration signal under the target working condition according to the diagnostic model and output the diagnostic results.

[0008] Preferably, a one-dimensional deformable convolutional network is used as the core module of the feature extractor. By dynamically adjusting the sampling position and weight of the convolutional kernel, the local temporal features in the vibration signal are adaptively captured, thereby completing the feature extraction of the vibration signal of the gearbox of the hoisting mechanism of the mining shovel robot under different working conditions.

[0009] Preferably, the two-stage domain adaptive strategy for fusing triple loss includes: A diagnostic model is trained on source domain data, and active learning is performed using the diagnostic model. The samples with the lowest or highest prediction confidence are selected from the target domain samples for expert annotation to obtain the labels of these samples. The labeled samples from the source domain and the target domain are then mixed and data augmentation based on linear interpolation is performed. Using triplet loss to guide the diagnostic model to learn domain-invariant features, we can achieve domain-adaptive fault diagnosis across operating conditions.

[0010] Preferably, in the two-stage domain adaptive strategy that incorporates triple loss: The loss function used in the first stage is:

[0011] in Represents source domain samples, Represents cross-entropy loss, This indicates that cross-entropy loss is applied to samples in the source domain. Indicates the loss of the triplet. This indicates that triplet loss is applied to the source domain samples. Indicates the loss weight; The loss function used in the second stage is:

[0012]

[0013] in, This represents the target domain sample, in its initial state. The samples in the sample list are all unlabeled. This represents the set of labeled target domain samples obtained by expert annotation after applying an active learning strategy to unlabeled target domain samples; Indicates to and Perform a union operation first, then perform data augmentation operations; data augmentation operations The specific operation method is as follows As shown, where This means randomly selecting two different samples from the set being operated on, and then... The new sample is obtained by performing linear interpolation. Then, perform a union operation on all the newly generated sample sets and the original set, which is a data augmentation operation. That is, it is completed; This means that for samples that are still unlabeled after adopting the active learning strategy, pseudo-labels are used, and the corresponding cross-entropy loss is obtained by calculating the pseudo-labels; Both represent the loss weight.

[0014] Preferably, a fault diagnosis model for the gearbox of the hoisting mechanism of a mining shovel robot is constructed and trained using the nn package under the PyTorch framework, and the model performance is evaluated by comparing the diagnosis results with the actual situation.

[0015] A cross-condition fault diagnosis system for the gearbox of a mining shovel robot provided by the present invention includes: Module M1: Collects vibration signals from the gearbox of the hoisting mechanism of a mining shovel robot under different working conditions; Module M2: Constructs a diagnostic network and performs adaptive feature extraction on the vibration signal; The diagnostic network includes a feature extractor and a fault classifier from a one-dimensional deformable convolutional network. Module M3: Based on the extracted features, the diagnostic network is trained using a two-stage domain adaptive strategy that incorporates triplet loss, and the diagnostic model is output. Module M4: Performs fault diagnosis on the vibration signal under the target working condition based on the diagnostic model and outputs the diagnostic results.

[0016] Preferably, a one-dimensional deformable convolutional network is used as the core module of the feature extractor. By dynamically adjusting the sampling position and weight of the convolutional kernel, the local temporal features in the vibration signal are adaptively captured, thereby completing the feature extraction of the vibration signal of the gearbox of the hoisting mechanism of the mining shovel robot under different working conditions.

[0017] Preferably, the two-stage domain adaptive strategy for fusing triple loss includes: A diagnostic model is trained on source domain data, and active learning is performed using the diagnostic model. The samples with the lowest or highest prediction confidence are selected from the target domain samples for expert annotation to obtain the labels of these samples. The labeled samples from the source domain and the target domain are then mixed and data augmentation based on linear interpolation is performed. Using triplet loss to guide the diagnostic model to learn domain-invariant features, we can achieve domain-adaptive fault diagnosis across operating conditions.

[0018] Preferably, in the two-stage domain adaptive strategy that incorporates triple loss: The loss function used in the first stage is:

[0019] in Represents source domain samples, Represents cross-entropy loss, This indicates that cross-entropy loss is applied to samples in the source domain. Indicates the loss of the triplet. This indicates that triplet loss is applied to the source domain samples. Indicates the loss weight; The loss function used in the second stage is:

[0020]

[0021] in, This represents the target domain sample, in its initial state. The samples in the sample list are all unlabeled. This represents the set of labeled target domain samples obtained by expert annotation after applying an active learning strategy to unlabeled target domain samples; Indicates to and Perform a union operation first, then perform data augmentation operations; data augmentation operations The specific operation method is as follows As shown, where This means randomly selecting two different samples from the set being operated on, and then... The new sample is obtained by performing linear interpolation. Then, perform a union operation on all the newly generated sample sets and the original set, which is a data augmentation operation. That is, it is completed; This means that for samples that are still unlabeled after adopting the active learning strategy, pseudo-labels are used, and the corresponding cross-entropy loss is obtained by calculating the pseudo-labels; Both represent the loss weight.

[0022] Preferably, a fault diagnosis model for the gearbox of the hoisting mechanism of a mining shovel robot is constructed and trained using the nn package under the PyTorch framework, and the model performance is evaluated by comparing the diagnosis results with the actual situation.

[0023] Compared with the prior art, the present invention has the following beneficial effects: 1. This invention introduces a one-dimensional deformable convolutional network to adaptively adjust the sampling position and weight of the convolutional kernel, effectively capturing vibration signal features under different speeds, torques and operating conditions, significantly enhancing the model's feature extraction capability under varying operating conditions, and overcoming the performance bottleneck of traditional convolutional networks under non-uniform sampling and temporal variations.

[0024] 2. This invention utilizes triplet loss to optimize the feature space during the source domain training phase, enhancing intra-class compactness and inter-class separability, enabling the model to learn more discriminative feature representations, improving the accuracy and stability of fault classification, and maintaining high diagnostic accuracy even under weak signal and noise interference.

[0025] 3. This invention combines an active learning mechanism to dynamically select the most informative samples for expert annotation by predicting confidence levels, which significantly reduces the reliance on a large amount of labeled data, lowers the cost of manual annotation, and maximizes performance under a limited annotation budget. It is suitable for solving the problem of difficult data annotation in real industrial scenarios.

[0026] 4. This invention adopts a two-stage domain adaptive strategy. In the first stage, feature optimization is performed on the source domain. In the second stage, data augmentation and domain-invariant feature learning are performed by mixing source and target domain labeled samples. This effectively alleviates the domain shift problem caused by changes in working conditions and achieves efficient fault diagnosis across working conditions. Attached Figure Description

[0027] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings: Figure 1 This is a flowchart illustrating the implementation of the cross-working-condition fault diagnosis method for the gearbox of the hoisting mechanism of the mining shovel robot proposed in this invention.

[0028] Figure 2 This is a flowchart illustrating the implementation of the one-dimensional deformable convolution method proposed in this invention.

[0029] Figure 3 This is a flowchart of the method of the present invention. Detailed Implementation

[0030] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0031] like Figure 3 As shown, a method for cross-working-condition fault diagnosis of gearbox in a mining shovel robot includes: Step S1: Collect vibration signals of the gearbox of the hoisting mechanism of the mining shovel robot under different working conditions; Step S2: Construct a diagnostic network and perform adaptive feature extraction on the vibration signal; The diagnostic network includes a feature extractor and a fault classifier from a one-dimensional deformable convolutional network. Step S3: Based on the extracted features, the diagnostic network is trained using a two-stage domain adaptive strategy that incorporates triplet loss, and the diagnostic model is output. Step S4: Perform fault diagnosis on the vibration signal under the target working condition according to the diagnostic model and output the diagnostic results.

[0032] In this invention, a fault diagnosis model for the gearbox of a mining shovel-loading robot's hoisting mechanism achieves efficient cross-condition fault diagnosis under limited labeled samples by integrating deformable convolution, triplet loss, and active learning mechanisms. The model takes vibration signals as input and employs a one-dimensional deformable convolutional network as a feature extractor. It dynamically adjusts the sampling position and weights of the convolutional kernels based on the signal's time-domain waveform and frequency-domain energy distribution, thereby adaptively capturing key fault features in the gearbox vibration signals under different operating conditions, significantly improving the adaptability and robustness of feature extraction. In the first stage, the source domain data is trained using triplet loss, enabling similar fault samples to form compact clusters in the feature space, maintaining a large distance between different categories, enhancing the discriminative power of feature representation, and providing a structured feature foundation for subsequent domain adaptation. In the second stage, based on the current classification results, the model actively selects samples with the highest and lowest prediction confidence in the target domain for expert annotation: high-confidence samples reflect the model's strong recognition ability for certain categories, but may lead to systematic misjudgments due to domain bias; low-confidence samples are near the decision boundary, reflecting the model's uncertainty under cross-operating conditions. By annotating these samples, the model can effectively identify and correct systematic biases, while strengthening its ability to identify reliable predictions. The annotated samples are mixed with the source domain data, and data augmentation is performed through linear interpolation to generate synthetic samples with features from both the source and target domains, further promoting the fusion of features from the two domains. On this basis, triplet loss is continuously applied to the mixed samples, guiding the model to learn cross-domain consistent feature representations, thereby achieving effective alignment between the source and target domains. Ultimately, this model achieves intelligent and reliable fault diagnosis of the gearbox of the hoisting mechanism of a mining loader robot under various operating conditions without relying on additional sensors or simulation data.

[0033] Example 1 like Figure 1 As shown, a method for cross-working-condition fault diagnosis of gearbox in a mining shovel robot includes: Step S1: Collect vibration signals of the gearbox of the hoisting mechanism of the mining loader robot under typical mining conditions online. The collected conditions cover different loads, speeds, accelerations and decelerations, and start-stop operations, covering different operating rhythms of the gearbox of the hoisting mechanism of the mining loader robot, as well as typical operating states with frequent start-stop operations and impact loads.

[0034] Step S2: As Figure 2As shown, a diagnostic network employing a one-dimensional deformable convolutional network (BDCNN) is constructed, integrating an adaptive feature extractor and a fault classifier. The BDCNN, based on the time-domain and frequency-domain characteristics of the input signal, utilizes the stochastic gradient descent algorithm in neural networks to adaptively adjust the sampling position of the convolutional kernel according to different inputs while learning the kernel weights, thereby achieving better feature extraction and realizing adaptive feature extraction of vibration signals. This adaptive mechanism enables the model to adapt to the non-stationarity and frequency shift characteristics of vibration signals under different working conditions and rhythms in actual mines, improving the robustness of feature extraction.

[0035] Step S3: Design a two-stage domain adaptation strategy incorporating triplet loss. In the first stage, the model is trained on the source domain data, and triplet loss is used to enhance intra-class compactness and inter-class separability, thereby enhancing sample representativeness and laying a solid foundation for active learning and domain adaptation in the second stage. In the second stage, the model obtained in the first stage is used for active learning. Samples with the lowest or highest prediction confidence are selected from the target domain samples for expert annotation, obtaining labels for these samples. Then, the labeled samples from the source and target domains are mixed and data augmentation based on linear interpolation is performed. Finally, in addition to ordinary cross-entropy loss, triplet loss is used to guide the model to learn domain-invariant features, thereby achieving domain-adaptive fault diagnosis across operating conditions. Step S4: Use the nn package under the PyTorch framework to build a fault diagnosis model of the gearbox of the hoisting mechanism of the mining shovel robot and train it, and evaluate the model performance by comparing the diagnosis results with the actual situation.

[0036] The fault diagnosis model for the gearbox of the hoisting mechanism of the mining shovel robot integrates deformable convolution, triplet loss, and active learning, effectively improving the model's cross-condition fault diagnosis capability with limited labeled samples. First, deformable convolution automatically selects the optimal convolution kernel weights and sampling positions based on the input signal, naturally suited for feature extraction of vibration signals under complex mining conditions. Second, the triplet loss in the first-stage domain adaptation effectively enhances the representativeness of the samples, laying the foundation for the second stage. The active learning in the second stage, through expert annotation of high-confidence samples (reflecting the model's strong recognition ability for specific categories, but potentially implying systematic misjudgments) and low-confidence samples (reflecting model uncertainty, mostly easily confused samples under cross-conditions), effectively corrects the model's systematic bias and strengthens the ability to identify reliable predictions. Simultaneously, data augmentation based on linear interpolation implicitly promotes the fusion between the source and target domains. Based on this, triplet loss is used to guide the model to learn domain-invariant features, ultimately achieving efficient cross-condition domain adaptation. In summary, the model achieves intelligent fault diagnosis of the gearbox of the hoisting mechanism of a mining shovel robot under complex mining conditions.

[0037] In step S2, a one-dimensional deformable convolutional network is used as the core module of the feature extractor. By dynamically adjusting the sampling position and weight of the convolutional kernel, the local temporal features in the vibration signal are adaptively captured, thereby achieving efficient feature extraction of the vibration signal of the gearbox of the hoisting mechanism of the mining shovel robot under different working conditions.

[0038] In step S3, a two-stage transfer learning strategy is adopted. In the first stage, the triplet loss is used to optimize the feature space in the source domain training stage, enhance intra-class compactness and inter-class separability, thereby improving the feature representation ability and classification performance of the model in the cross-condition transfer process in the second stage.

[0039] The loss function used in the first stage is:

[0040] in Represents source domain samples, Represents cross-entropy loss, This indicates that cross-entropy loss is applied to samples in the source domain. Indicates the loss of the triplet. This indicates that triplet loss is applied to the source domain samples. Indicates the loss weight; The loss function used in the second stage is:

[0041]

[0042] in, This represents the target domain sample, in its initial state. The samples in the sample list are all unlabeled. This represents the set of labeled target domain samples obtained by expert annotation after applying an active learning strategy to unlabeled target domain samples; Indicates to and Perform a union operation first, then perform data augmentation operations; data augmentation operations The specific operation method is as follows As shown, where This means randomly selecting two different samples from the set being operated on, and then... The new sample is obtained by performing linear interpolation. Then, perform a union operation on all the newly generated sample sets and the original set, which is a data augmentation operation. That is, it is completed; This means that for samples that are still unlabeled after adopting the active learning strategy, pseudo-labels are used. The corresponding cross-entropy loss is obtained by calculating the pseudo-labels (i.e., the most likely fault types as perceived by the diagnostic model). Both represent the loss weight.

[0043] The proposed model was tested and evaluated, and it demonstrated good generalization performance in transfer tasks across different working conditions. High diagnostic accuracy was achieved in multiple cross-working-condition scenarios. In the transfer learning fault diagnosis experiment on the gearbox of the hoisting mechanism of a mining shovel robot under different speeds, torques, and operating states, with a 12.5% ​​label budget, the average fault diagnosis accuracy reached 96.27%, 95.86%, and 96.06%, respectively. This indicates that the proposed fault diagnosis model for the gearbox of the hoisting mechanism of the mining shovel robot has strong adaptability in transfer tasks across different speeds, torques, and operating states (such as acceleration and deceleration), and can effectively address the feature shift problem caused by changes in working conditions.

[0044] The above results demonstrate that the proposed model integrates deformable convolution, triplet loss, and active learning, significantly improving the model's diagnostic robustness and generalization ability under various operating conditions. It can achieve high-precision, low-label-dependency intelligent fault diagnosis, providing reliable technical support for online monitoring and intelligent operation and maintenance of industrial equipment.

[0045] The present invention also provides a fault diagnosis system for the gearbox of a mining shovel-loading robot under various working conditions. The fault diagnosis system for the gearbox of a mining shovel-loading robot under various working conditions can be implemented by executing the process steps of the fault diagnosis method for the gearbox of a mining shovel-loading robot under various working conditions. That is, those skilled in the art can understand the fault diagnosis method for the gearbox of a mining shovel-loading robot under various working conditions as a preferred embodiment of the fault diagnosis system for the gearbox of a mining shovel-loading robot under various working conditions.

[0046] Specifically, a cross-condition fault diagnosis system for the gearbox of a mining shovel loading robot includes: Module M1: Collects vibration signals from the gearbox of the hoisting mechanism of a mining shovel robot under different working conditions; Module M2: Constructs a diagnostic network and performs adaptive feature extraction on the vibration signal; The diagnostic network includes a feature extractor and a fault classifier from a one-dimensional deformable convolutional network. Module M3: Based on the extracted features, the diagnostic network is trained using a two-stage domain adaptive strategy that incorporates triplet loss, and the diagnostic model is output. Module M4: Performs fault diagnosis on the vibration signal under the target working condition based on the diagnostic model and outputs the diagnostic results.

[0047] A one-dimensional deformable convolutional network is used as the core module of the feature extractor. By dynamically adjusting the sampling position and weight of the convolutional kernel, the local temporal features in the vibration signal are adaptively captured, thus completing the feature extraction of the vibration signal of the gearbox of the hoisting mechanism of the mining shovel robot under different working conditions.

[0048] The two-stage domain adaptive strategy that integrates triple loss includes: A diagnostic model is trained on source domain data, and active learning is performed using the diagnostic model. The samples with the lowest or highest prediction confidence are selected from the target domain samples for expert annotation to obtain the labels of these samples. The labeled samples from the source domain and the target domain are then mixed and data augmentation based on linear interpolation is performed. Using triplet loss to guide the diagnostic model to learn domain-invariant features, we can achieve domain-adaptive fault diagnosis across operating conditions.

[0049] In the two-stage domain adaptive strategy that incorporates triple loss: The loss function used in the first stage is:

[0050] in Represents source domain samples, Represents cross-entropy loss, This indicates that cross-entropy loss is applied to samples in the source domain. Indicates the loss of the triplet. This indicates that triplet loss is applied to the source domain samples. Indicates the loss weight; The loss function used in the second stage is:

[0051]

[0052] in, This represents the target domain sample, in its initial state. The samples in the sample list are all unlabeled. This represents the set of labeled target domain samples obtained by expert annotation after applying an active learning strategy to unlabeled target domain samples; Indicates to and Perform a union operation first, then perform data augmentation operations; data augmentation operations The specific operation method is as follows As shown, where This means randomly selecting two different samples from the set being operated on, and then... The new sample is obtained by performing linear interpolation. Then, perform a union operation on all the newly generated sample sets and the original set, which is a data augmentation operation. That is, it is completed; This means that for samples that are still unlabeled after adopting the active learning strategy, pseudo-labels are used. The corresponding cross-entropy loss is obtained by calculating the pseudo-labels (i.e., the most likely fault types as perceived by the diagnostic model). Both represent the loss weight.

[0053] A fault diagnosis model for the gearbox of a mining shovel-loading robot hoisting mechanism was built and trained using the nn package under the PyTorch framework. The model performance was then evaluated by comparing the diagnostic results with actual conditions.

[0054] Those skilled in the art will understand that, besides implementing the system and its various devices, modules, and units provided by this invention in the form of purely computer-readable program code, the same functions can be achieved entirely through logical programming of the method steps, making the system and its various devices, modules, and units of this invention function in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers, and embedded microcontrollers. Therefore, the system and its various devices, modules, and units provided by this invention can be considered as a hardware component, and the devices, modules, and units included therein for implementing various functions can also be considered as structures within the hardware component; alternatively, the devices, modules, and units for implementing various functions can be considered as both software modules implementing the method and structures within the hardware component.

[0055] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A method for cross-working-condition fault diagnosis of gearbox in a mining shovel loading robot, characterized in that, include: Step S1: Collect vibration signals of the gearbox of the hoisting mechanism of the mining shovel robot under different working conditions; Step S2: Construct a diagnostic network and perform adaptive feature extraction on the vibration signal; The diagnostic network includes a feature extractor and a fault classifier from a one-dimensional deformable convolutional network. Step S3: Based on the extracted features, the diagnostic network is trained using a two-stage domain adaptive strategy that incorporates triplet loss, and the diagnostic model is output. Step S4: Perform fault diagnosis on the vibration signal under the target working condition according to the diagnostic model and output the diagnostic results.

2. The method for cross-working-condition fault diagnosis of the gearbox of a mining shovel loading robot according to claim 1, characterized in that, A one-dimensional deformable convolutional network is used as the core module of the feature extractor. By dynamically adjusting the sampling position and weight of the convolutional kernel, the local temporal features in the vibration signal are adaptively captured, thus completing the feature extraction of the vibration signal of the gearbox of the hoisting mechanism of the mining shovel robot under different working conditions.

3. The method for cross-working-condition fault diagnosis of the gearbox of a mining shovel loading robot according to claim 1, characterized in that, The two-stage domain adaptive strategy that integrates triple loss includes: A diagnostic model is trained on source domain data, and active learning is performed using the diagnostic model. The samples with the lowest or highest prediction confidence are selected from the target domain samples for expert annotation to obtain the labels of these samples. The labeled samples from the source domain and the target domain are then mixed and data augmentation based on linear interpolation is performed. Using triplet loss to guide the diagnostic model to learn domain-invariant features, we can achieve domain-adaptive fault diagnosis across operating conditions.

4. The method for cross-working-condition fault diagnosis of the gearbox of a mining shovel loading robot according to claim 3, characterized in that, In the two-stage domain adaptive strategy that incorporates triple loss: The loss function used in the first stage is: in Represents source domain samples, Represents cross-entropy loss, This indicates that cross-entropy loss is applied to samples in the source domain. Indicates the loss of the triplet. This indicates that triplet loss is applied to the source domain samples. Indicates the loss weight; The loss function used in the second stage is: in, This represents the target domain sample, in its initial state. The samples in the sample list are all unlabeled. This represents the set of labeled target domain samples obtained by expert annotation after applying an active learning strategy to unlabeled target domain samples; Indicates to and Perform a union operation first, then perform data augmentation operations; data augmentation operations The specific operation method is as follows As shown, where This means randomly selecting two different samples from the set being operated on, and then... The new sample is obtained by performing linear interpolation. Then, perform a union operation on all the newly generated sample sets and the original set, which is a data augmentation operation. That is, it is completed; This means that for samples that are still unlabeled after adopting the active learning strategy, pseudo-labels are used, and the corresponding cross-entropy loss is obtained by calculating the pseudo-labels; Both represent the loss weight.

5. The method for cross-working-condition fault diagnosis of the gearbox of a mining shovel loading robot according to claim 1, characterized in that, A fault diagnosis model for the gearbox of a mining shovel-loading robot hoisting mechanism was built and trained using the nn package under the PyTorch framework. The model performance was then evaluated by comparing the diagnostic results with actual conditions.

6. A cross-working-condition fault diagnosis system for the gearbox of a mining shovel loading robot, characterized in that, include: Module M1: Collects vibration signals from the gearbox of the hoisting mechanism of a mining shovel robot under different working conditions; Module M2: Constructs a diagnostic network and performs adaptive feature extraction on the vibration signal; The diagnostic network includes a feature extractor and a fault classifier from a one-dimensional deformable convolutional network. Module M3: Based on the extracted features, the diagnostic network is trained using a two-stage domain adaptive strategy that incorporates triplet loss, and the diagnostic model is output. Module M4: Performs fault diagnosis on the vibration signal under the target working condition based on the diagnostic model and outputs the diagnostic results.

7. The cross-condition fault diagnosis system for the gearbox of a mining shovel loading robot according to claim 6, characterized in that, A one-dimensional deformable convolutional network is used as the core module of the feature extractor. By dynamically adjusting the sampling position and weight of the convolutional kernel, the local temporal features in the vibration signal are adaptively captured, thus completing the feature extraction of the vibration signal of the gearbox of the hoisting mechanism of the mining shovel robot under different working conditions.

8. The cross-condition fault diagnosis system for the gearbox of a mining shovel loading robot according to claim 6, characterized in that, The two-stage domain adaptive strategy that integrates triple loss includes: A diagnostic model is trained on source domain data, and active learning is performed using the diagnostic model. The samples with the lowest or highest prediction confidence are selected from the target domain samples for expert annotation to obtain the labels of these samples. The labeled samples from the source domain and the target domain are then mixed and data augmentation based on linear interpolation is performed. Using triplet loss to guide the diagnostic model to learn domain-invariant features, we can achieve domain-adaptive fault diagnosis across operating conditions.

9. The cross-condition fault diagnosis system for the gearbox of a mining shovel loading robot according to claim 8, characterized in that, In the two-stage domain adaptive strategy that incorporates triple loss: The loss function used in the first stage is: in Represents source domain samples, Represents cross-entropy loss, This indicates that cross-entropy loss is applied to samples in the source domain. Indicates the loss of the triplet. This indicates that triplet loss is applied to the source domain samples. Indicates the loss weight; The loss function used in the second stage is: in, This represents the target domain sample, in its initial state. The samples in the sample list are all unlabeled. This represents the set of labeled target domain samples obtained by expert annotation after applying an active learning strategy to unlabeled target domain samples; Indicates to and Perform a union operation first, then perform data augmentation operations; data augmentation operations The specific operation method is as follows As shown, where This means randomly selecting two different samples from the set being operated on, and then... The new sample is obtained by performing linear interpolation. Then, perform a union operation on all the newly generated sample sets and the original set, which is a data augmentation operation. That is, it is completed; This means that for samples that are still unlabeled after adopting the active learning strategy, pseudo-labels are used, and the corresponding cross-entropy loss is obtained by calculating the pseudo-labels; Both represent the loss weight.

10. The cross-condition fault diagnosis system for the gearbox of a mining shovel loading robot according to claim 6, characterized in that, A fault diagnosis model for the gearbox of a mining shovel-loading robot hoisting mechanism was built and trained using the nn package under the PyTorch framework. The model performance was then evaluated by comparing the diagnostic results with actual conditions.

Images