Emergency stop protection control system and method for multi-working-condition gearbox test bed

Abstract

The invention belongs to the technical field of fingerprint identification, and provides an emergency stop protection control system and method for a multi-working-condition gearbox test bed, and the method comprises the steps: carrying out the timestamp alignment and time sequence feature processing of collected data, extracting the RMS, peak-to-peak value and kurtosis coefficient of a vibration signal based on the time sequence feature processing, and recognizing the fault type; the running state of the gear is analyzed through FFT, and bearing faults are analyzed through an envelope spectrum; a space-time attention network model is adopted, historical test bed data is used for pre-training, Gaussian noise and a virtual fault sample are injected during adversarial training to improve the robustness of the model, and material mechanics constraints are embedded into a dynamic threshold value output by the model; threshold value verification and adjustment: a reinforcement learning model based on a deep Q network is used as a discrimination model architecture, and an intelligent decision system is constructed to verify and adjust the threshold value of the multi-working-condition gearbox test bench; and according to the predicted dynamic threshold result, a mechanical braking system is adopted to realize braking.

Description

Technical Field

[0001] The invention belongs to the technical field of gearbox testing equipment, and in particular to an emergency stop protection control system and method for a multi-working-condition gearbox test bench. Background Art

[0002] With the development of modern industry, gearboxes are widely used in various mechanical equipment. In order to ensure the reliability and safety of gearboxes under different working conditions, comprehensive performance testing is required. The multi-working condition gearbox test bench is an important equipment for simulating the actual working environment of the gearbox. The performance of its emergency stop protection control system is directly related to the safe conduct of the test and the service life of the equipment.

[0003] Traditional emergency stop systems have many flaws, such as the problem of threshold fixation. The fixed protection threshold cannot adapt to dynamic operating conditions such as variable speed and variable load, resulting in high false alarm and missed alarm rates. Moreover, the existing system analyzes vibration, temperature, and torque parameters independently and lacks the ability to identify failure modes through multi-physics field coupling. These problems have seriously affected the operating efficiency and safety of multi-operating condition gearbox test benches and need to be addressed urgently.

[0004] To this end, the present invention provides an emergency stop protection control system and method for a multi-operating-condition gearbox test bench. Summary of the Invention

[0005] In order to make up for the deficiencies of the prior art, at least one technical problem raised in the background technology is solved.

[0006] The technical solution adopted by the present invention to solve its technical problem is: In a first aspect, the present invention provides an emergency stop protection control method for a multi-operating-condition gearbox test bench, comprising: Step 1: Real-time data collection of the gearbox during operation; Step 2: Align the collected data with timestamps and perform time series feature processing. Based on this time series feature processing, extract the RMS, peak-to-peak value, and kurtosis coefficient of the vibration signal and identify the fault type. Simultaneously, use FFT to analyze the operating status of the gears, and use envelope spectrum to analyze bearing faults. Step 3: Dynamic threshold generation model training. This model uses a spatiotemporal attention network model and is pre-trained using historical testbed data. During adversarial training, Gaussian noise and virtual fault samples are injected to improve the model's robustness. Material mechanics constraints are embedded in the model's output dynamic threshold. Step 4: Threshold verification and adjustment: Using a reinforcement learning model based on a deep Q network as the discriminant model architecture, an intelligent decision-making system is built to verify and adjust the threshold of the multi-condition gearbox test bench. Step 5: Based on the predicted dynamic threshold results, braking is achieved using a mechanical braking system.

[0007] In a second aspect, the present invention provides an emergency stop protection control system for a multi-operating-condition gearbox test bench, comprising: Data acquisition module: collects data of the gearbox in real time during operation; Data processing module: This module performs time stamp alignment and time series feature processing on the collected data. Based on this processing, it extracts the RMS, peak-to-peak value, and kurtosis coefficient of the vibration signal and identifies the fault type. It also uses FFT to analyze the operating status of the gears and envelope spectrum to analyze bearing faults. Dynamic Threshold Generation Module: This module trains a dynamic threshold generation model using a spatiotemporal attention network model and pre-training with historical test bench data. During adversarial training, Gaussian noise and virtual fault samples are injected to improve the model's robustness. Material mechanics constraints are embedded in the model's output dynamic threshold. Dynamic threshold adjustment module: threshold verification and adjustment, using a reinforcement learning model based on a deep Q network as the discriminant model architecture, and building an intelligent decision-making system to verify and adjust the threshold of a multi-condition gearbox test bench; Control module: Based on the predicted dynamic threshold results, the mechanical braking system is used to achieve braking.

[0008] The beneficial effects of the present invention are as follows: The spatiotemporal attention network model effectively captures the dynamic correlation between operating parameters and fault characteristics. It captures parameter trends over time in the temporal dimension and highlights fault-related sensor data features in the spatial dimension, thereby generating dynamic thresholds that accurately reflect reasonable thresholds under different operating conditions. Compared to traditional fixed thresholds, this model better adapts to the complex and changing operating conditions of gearboxes, reducing false positives and missed alarms. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] The present invention will be further described below with reference to the accompanying drawings.

[0010] Figure 1 is a flowchart of the steps of Example 1 of the present invention; Figure 2 This is a module diagram of Example 2 of the present invention. DETAILED DESCRIPTION

[0011] In order to make the technical means, creative features, objectives and effects achieved by the present invention easier to understand, the present invention is further described below in conjunction with specific implementation methods. Example 1

[0012] like Figure 1 As shown, a multi-operating-condition gearbox test bench emergency stop protection control method according to an embodiment of the present invention includes the following steps: Step 1: Real-time data collection of the gearbox during operation; Data acquisition, installation of speed sensors, load sensors, vibration sensors, and temperature sensors; these sensors are installed in key locations of the gearbox and test bench to collect real-time speed, load, vibration, and temperature parameters of the gearbox during operation; Step 2: Align the collected data with timestamps and perform time series feature processing. Based on this time series feature processing, extract the RMS, peak-to-peak value, and kurtosis coefficient of the vibration signal and identify the fault type. Simultaneously, use FFT to analyze the operating status of the gears, and use envelope spectrum to analyze bearing faults. Multi-data fusion and processing: In a multi-operating gearbox test bench, different types of sensors often have different sampling rates due to their own characteristics and hardware design differences. To achieve spatiotemporal synchronization and ensure timestamp alignment accuracy, a combination of hardware and software can be used. At the hardware level, a high-precision clock synchronization module is selected, and clock synchronization equipment based on the Global Positioning System (GPS) is used to provide a unified time reference signal for all sensors. When each sensor collects data, it embeds this precise timestamp into the data frame. In terms of software, data interpolation and resampling algorithms are used to process data with different sampling rates. For data with lower sampling rates, linear interpolation and spline interpolation methods are used to supplement missing data points on the time axis based on existing data points, thereby improving the temporal resolution of the data. For data with excessively high sampling rates, downsampling is performed to reduce the data volume while ensuring that key data features are not lost, ultimately ensuring that all sensor data is strictly aligned in time. Based on time series feature engineering, the RMS, peak-to-peak, and kurtosis coefficients of the vibration signal are first extracted. FFT and envelope spectrum analysis are then used to identify the fault characteristic frequency. During the test, the speed gear position of the gearbox during operation is then marked. In step 2, the RMS value of the vibration signal can effectively reflect the energy intensity of the signal. It is obtained by averaging the square values ​​of the vibration signal over a period of time and then taking the square root. The calculation formula is: , where N is the number of signal sampling points. In practical applications, when the gearbox operates normally, the vibration RMS value is in a relatively stable range; if the gear is worn or pitting, the vibration energy will increase and the RMS value will rise accordingly. This can be used as an important indicator for early warning of faults. is the discrete vibration signal value; The peak-to-peak value is the difference between the maximum and minimum values ​​of the vibration signal within a cycle. It directly reflects the severity of the vibration. For gearboxes, when a serious fault such as a broken gear occurs, the peak-to-peak value of the vibration will increase dramatically, far exceeding the normal operating range. By continuously monitoring the peak-to-peak value changes, such sudden and serious fault conditions can be discovered in a timely manner. The kurtosis coefficient is used to measure the impact characteristics of the vibration signal. The calculation formula is: ;in, The kurtosis coefficient of the vibration signal of a normally operating gearbox usually fluctuates within a specific range. When a gear has a broken tooth or other fault, a strong impact is generated, causing the kurtosis value to increase significantly. Generally, when a gear has a broken tooth, the kurtosis value is greater than 5. This characteristic can be used to identify such faults. In step 2, specifically, in gearbox fault diagnosis, faults in different components will generate vibration responses of specific frequencies. For example, by performing FFT analysis on the vibration signal, the gear meshing frequency and its frequency harmonics can be identified. If the gear is worn, the amplitude at the meshing frequency will change, and sideband frequencies will appear. By observing the changes in these frequency characteristics, the operating status of the gear can be determined. Envelope spectrum analysis is mainly used to extract modulation information from signals and plays an important role in identifying bearing faults. In actual operation, bearing faults (such as outer ring faults) will cause periodic impact modulation of the vibration signal. By performing envelope demodulation on the vibration signal and then performing spectrum analysis, the envelope spectrum can be obtained. In the envelope spectrum, the frequency of the bearing outer ring fault can be clearly identified. Equal characteristic frequencies; the calculation formula for the bearing outer ring fault frequency is: ; Where M is the number of bearing rolling elements, is the rotation frequency of the shaft, d is the rolling element diameter, D is the bearing pitch diameter, is the contact angle. Based on the calculated theoretical fault frequency and the envelope spectrum analysis results, it can be accurately determined whether the bearing has an outer ring fault. During the test, the speed gear of the gearbox must be clearly marked. Speed ​​information can be obtained through a speed sensor installed on the input or output shaft of the gearbox. In the data acquisition system, the real-time speed collected by the speed sensor is compared with the preset speed gear threshold. When the speed is within the allowable error range of a certain gear, the corresponding speed gear label is marked for all data collected during that time period. For example, if the tolerance of the preset 800 rpm gear is ±50 rpm, when the speed sensor measures a speed between 750-850 rpm, the relevant data is marked as the 800 rpm gear; Load types are divided into constant torque and impact loads. The load type can be determined by a load sensor installed at the connection between the loading device and the gearbox. For constant torque loads, the signal output by the load sensor remains relatively stable over a period of time, with fluctuations within a very small range. However, impact loads can cause the load sensor signal to fluctuate significantly and instantaneously. During data processing, a threshold is set for determining impact loads. When the fluctuation amplitude of the load signal exceeds this threshold, it is determined to be an impact load and the relevant data is labeled with an impact load label. If the signal fluctuation is within the threshold range and is relatively stable, it is labeled as a constant torque load label. Step 3: Dynamic threshold generation model training. This uses a spatiotemporal attention network model and pre-training with a large amount of historical testbed data. To improve the model's robustness, Gaussian noise and virtual fault samples are injected during adversarial training. Material mechanics constraints are embedded in the model's output dynamic thresholds to ensure that the generated thresholds are physically meaningful and engineering feasible. In step three, first, specifically, in the time dimension, the recurrent neural network (RNN) or its variant, the long short-term memory network (LSTM) component, can effectively capture the dynamic change trend of the operating parameters over time; For example, for the fluctuations of gearbox speed and load parameters at different times, STAN can learn the time dependence between parameters and analyze the delayed response patterns of parameters such as vibration and temperature during the speed increase process; In the spatial dimension, the attention mechanism can model the spatial distribution relationship of different types of sensor data at the same time. It can automatically assign weights to highlight sensor data features closely related to fault characteristics and weaken irrelevant or interfering data. For example, when a gearbox has a specific fault, such as when the vibration signal characteristics in a certain frequency band are highly correlated with the fault, STAN will assign a higher attention weight to the vibration data in that frequency band, thereby accurately capturing the complex dynamic correlation between operating parameters and fault characteristics, laying the foundation for generating accurate dynamic thresholds; In step three, the second specific step is to improve the robustness of the model by injecting Gaussian noise into the training data to simulate the noise environment such as sensor measurement errors and electromagnetic interference that may occur in real applications. During the training process, the model continuously learns to accurately extract effective features under noise interference, thereby enhancing its resistance to noise and improving the model's stability in complex real-world environments. For example, the signal-to-noise ratio is set to 20 dB, that is, the ratio of the noise signal power to the original signal power is 1:100; through a large amount of historical test bench data, such as vibration, temperature, speed, and load data, random noise that conforms to the Gaussian distribution is superimposed; The SMOTE algorithm is used to generate virtual fault samples. In gearbox fault data, some fault types may have a small number of samples, resulting in insufficient model learning of these fault characteristics. The SMOTE algorithm generates new virtual fault samples by interpolating in the feature space of minority class fault samples. For example, for a relatively rare fault type like gear tooth breakage, the SMOTE algorithm generates new virtual samples based on the characteristics of existing broken tooth fault samples, increasing the number of broken tooth fault samples. These virtual fault samples are used together with the original fault samples and normal samples for model training, enabling the model to better learn various fault characteristics, especially rare fault characteristics, further improving the model's robustness and fault recognition capabilities. In step three, the second specific one is to embed material mechanics constraints into the dynamic threshold generation formula to ensure that the generated threshold has physical meaning and engineering feasibility; For example, for the torque threshold Tth, its upper limit is set to Tth ≤ 0.8σs, where σs is the yield strength of the gearbox material. The yield strength of the material is a key indicator of its mechanical properties. When the torque exceeds a certain percentage of the material's yield strength, the gearbox may undergo plastic deformation or even fail. By incorporating this material mechanics constraint into the threshold generation model, the model automatically considers the material's load-bearing capacity when generating the torque threshold, avoiding the generation of unreasonable thresholds. During the actual training process, the material mechanics constraint is optimized as part of the loss function. When the threshold generated by the model violates the constraint, the loss value is increased, prompting the model to adjust the parameters and generate a dynamic threshold that conforms to the material mechanics properties, thereby improving the reliability and safety of the model in actual engineering applications. Step 4: Threshold verification and adjustment. Using a reinforcement learning model based on a deep Q-network (DQN) as the discriminant model architecture, an intelligent decision-making system was built to verify and adjust the thresholds of the multi-condition gearbox test bench. Its state space design covered multiple key dimensions. In step 4, the first specific step is to record the threshold values ​​in the past period of time; These historical thresholds reflect the changing trends of the system's protection thresholds under different operating conditions. For example, under continuous high-speed, high-load conditions, the historical threshold sequence can show how the thresholds are dynamically adjusted as the conditions persist. By analyzing the historical threshold sequence, the DQN can learn the reasonable value range and adjustment mode of the thresholds under similar conditions, providing a reference for validating the current thresholds. Analyze the equipment health index, which combines the health status of the gearbox equipment reflected by vibration sensors, temperature sensors, and torque sensors; For example, an increase in vibration amplitude, an abnormal rise in temperature, or unstable torque fluctuations can all cause the device health index to decrease. As part of the state space, the device health index allows the DQN to take into account the actual operating conditions of the current device. When the device health index is low, it indicates that the gearbox may be on the verge of failure or has already failed. In this case, more careful verification and adjustment of the threshold are required to ensure the effectiveness of the emergency stop protection system. In step 4, the second specific example is that when the system detects five consecutive false alarms, it indicates that the currently set threshold may be too sensitive and needs to be adjusted appropriately. At this time, the threshold increase mechanism is activated with an increase of +Δ, where Δ is between 3% and 10%. The specific increase can be dynamically adjusted according to actual conditions. For example, if false alarms are frequent and seriously affect the test progress, a larger Δ value can be selected; if false alarms are relatively minor, a smaller Δ value can be used. In actual operation, the system will record relevant information about each false alarm, including the operating parameters, equipment status, and environmental conditions at the time. By analyzing these false alarm data and combining the DQN's learning results on the state space, the appropriate threshold increase range is determined. For example, under certain operating conditions, when continuous false alarms occur due to high ambient temperature, analysis shows that the temperature causes deviations in sensor signals, leading to false alarms. In this case, based on the threshold adjustment strategy learned by DQN in high-temperature environments, an appropriate Δ value is selected to increase the threshold to reduce the occurrence of false alarms. In step 4, the third specific example is that when the operating condition suddenly changes, such as when the speed step is greater than 30% / s, the operating state of the gearbox will change dramatically in a short period of time. The original threshold cannot adapt to this sudden change in time, and the dynamic compensation algorithm needs to be triggered. The Kalman filter algorithm is used here to predict threshold drift. Kalman filtering is an efficient recursive filtering algorithm that can optimally estimate the system state based on the system's state equation and observation equation. When the operating conditions suddenly change, the Kalman filter uses existing historical data and current observation data, such as speed, load, vibration, and temperature parameters, to predict the threshold drift. For example, when the speed suddenly rises sharply, the Kalman filter will predict the changes in parameters such as vibration and temperature that may be caused by the sudden change in speed based on the speed change rate and the changing trends of other related parameters, and then predict the direction and magnitude of the threshold drift. Based on the prediction results of the Kalman filter, the system dynamically compensates and adjusts the threshold, allowing the emergency stop protection system to adapt to the sudden change in operating conditions in a timely manner, effectively avoiding the spread of faults caused by threshold lag. During the adjustment process, the Kalman filter will continuously update the prediction results to adapt to the continuous changes in operating conditions, ensuring that the threshold is always within a reasonable and effective range. Step 5: Based on the predicted dynamic threshold results, an optimized mechanical braking system is used to achieve fast response braking; In step five, first, specifically, during the operation of the multi-condition gearbox test bench, the predicted dynamic threshold results are obtained through data processing and analysis, dynamic threshold generation model training, threshold verification and adjustment steps; When the actual operating parameters of the test bench, such as speed, load, vibration, and temperature, are compared with the predicted dynamic thresholds, if the actual parameters are determined to be outside the dynamic threshold range, it indicates that the gearbox is in a faulty or dangerous state. At this time, the central control module will decide to initiate braking measures based on this result; the central control module will use the mechanical braking system to implement braking; The technical solution of this embodiment is as follows: speed, load, vibration, and temperature sensors are installed at key locations of the gearbox and test bench for data collection. To address the differences in sampling rates among different sensors, a combination of software and hardware is used to achieve spatiotemporal synchronization and data alignment, and time-series feature engineering is performed. Next, a spatiotemporal attention network model is used in conjunction with historical test bench data to train a dynamic threshold generation model. During training, Gaussian noise is injected and virtual fault samples are generated to improve robustness, and material mechanics constraints are embedded to ensure reasonable thresholds. Threshold verification and adjustment are then performed using a reinforcement learning model based on a deep Q network. Based on state-space information such as historical thresholds and the equipment health index, a floating mechanism and a Kalman filter dynamic compensation algorithm are used for false alarms and sudden changes in operating conditions. Finally, when the actual operating parameters of the test bench exceed the dynamic threshold range, the central control module activates the mechanical braking system for braking, achieving rapid response braking, effectively improving the accuracy and reliability of the emergency stop protection of the multi-operating-condition gearbox test bench and ensuring the safe conduct of the test. Example 2

[0013] like Figure 2 As shown, based on Example 1, the present invention provides an emergency stop protection control system for a multi-operating-condition gearbox test bench, including the following modules: Data acquisition module: collects data of the gearbox in real time during operation; Data acquisition, installation of speed sensors, load sensors, vibration sensors, and temperature sensors; these sensors are installed in key locations of the gearbox and test bench to collect real-time speed, load, vibration, and temperature parameters of the gearbox during operation; Data processing module: This module performs time stamp alignment and time series feature processing on the collected data. Based on this processing, it extracts the RMS, peak-to-peak value, and kurtosis coefficient of the vibration signal and identifies the fault type. It also uses FFT to analyze the operating status of the gears and envelope spectrum to analyze bearing faults. Multi-data fusion and processing: In a multi-operating gearbox test bench, different types of sensors often have different sampling rates due to their own characteristics and hardware design differences. To achieve spatiotemporal synchronization and ensure timestamp alignment accuracy, a combination of hardware and software can be used. At the hardware level, a high-precision clock synchronization module is selected, and clock synchronization equipment based on the Global Positioning System (GPS) is used to provide a unified time reference signal for all sensors. When each sensor collects data, it embeds this precise timestamp into the data frame. In terms of software, data interpolation and resampling algorithms are used to process data with different sampling rates. For data with lower sampling rates, linear interpolation and spline interpolation methods are used to supplement missing data points on the time axis based on existing data points, thereby improving the temporal resolution of the data. For data with excessively high sampling rates, downsampling is performed to reduce the data volume while ensuring that key data features are not lost, ultimately ensuring that all sensor data is strictly aligned in time. Based on time series feature engineering, the RMS, peak-to-peak, and kurtosis coefficients of the vibration signal are first extracted. FFT and envelope spectrum analysis are then used to identify the fault characteristic frequency. During the test, the speed gear position of the gearbox during operation is then marked. The RMS value of the vibration signal can effectively reflect the energy intensity of the signal. It is obtained by averaging the square values ​​of the vibration signal over a period of time and then taking the square root. The calculation formula is: , where N is the number of signal sampling points. In practical applications, when the gearbox operates normally, the vibration RMS value is in a relatively stable range; if the gear is worn or pitting, the vibration energy will increase and the RMS value will rise accordingly. This can be used as an important indicator for early warning of faults. is the discrete vibration signal value; The peak-to-peak value is the difference between the maximum and minimum values ​​of the vibration signal within a cycle. It directly reflects the severity of the vibration. For gearboxes, when a serious fault such as a broken gear occurs, the peak-to-peak value of the vibration will increase dramatically, far exceeding the normal operating range. By continuously monitoring the peak-to-peak value changes, such sudden and serious fault conditions can be discovered in a timely manner. The kurtosis coefficient is used to measure the impact characteristics of the vibration signal. The calculation formula is: ;in, The kurtosis coefficient of the vibration signal of a normally operating gearbox usually fluctuates within a specific range. When a gear has a broken tooth or other fault, a strong impact is generated, causing the kurtosis value to increase significantly. Generally, when a gear has a broken tooth, the kurtosis value is greater than 5. This characteristic can be used to identify such faults. In gearbox fault diagnosis, faults in different components will generate vibration responses of specific frequencies. For example, by performing FFT analysis on the vibration signal, the gear meshing frequency and its frequency harmonics can be identified. If the gear is worn, the amplitude of the meshing frequency will change, and sideband frequencies will appear. By observing the changes in these frequency characteristics, the operating status of the gear can be determined. Envelope spectrum analysis is mainly used to extract modulation information from signals and plays an important role in identifying bearing faults. In actual operation, bearing faults (such as outer ring faults) will cause periodic impact modulation of the vibration signal. By performing envelope demodulation on the vibration signal and then performing spectrum analysis, the envelope spectrum can be obtained. In the envelope spectrum, the frequency of the bearing outer ring fault can be clearly identified. Equal characteristic frequencies; the calculation formula for the bearing outer ring fault frequency is: ; Where M is the number of bearing rolling elements, is the rotation frequency of the shaft, d is the rolling element diameter, D is the bearing pitch diameter, is the contact angle. Based on the calculated theoretical fault frequency and the envelope spectrum analysis results, it can be accurately determined whether the bearing has an outer ring fault. During the test, the speed gear of the gearbox must be clearly marked. Speed ​​information can be obtained through a speed sensor installed on the input or output shaft of the gearbox. In the data acquisition system, the real-time speed collected by the speed sensor is compared with the preset speed gear threshold. When the speed is within the allowable error range of a certain gear, the corresponding speed gear label is marked for all data collected during that time period. Load types are divided into constant torque and impact loads. The load type can be determined by a load sensor installed at the connection between the loading device and the gearbox. For constant torque loads, the signal output by the load sensor remains relatively stable over a period of time, with fluctuations within a very small range. However, impact loads can cause the load sensor signal to fluctuate significantly and instantaneously. During data processing, a threshold is set for determining impact loads. When the fluctuation amplitude of the load signal exceeds this threshold, it is determined to be an impact load and the relevant data is labeled with an impact load label. If the signal fluctuation is within the threshold range and is relatively stable, it is labeled as a constant torque load label. Dynamic Threshold Generation Module: This module trains a dynamic threshold generation model using a spatiotemporal attention network model and pre-training with historical test bench data. During adversarial training, Gaussian noise and virtual fault samples are injected to improve the model's robustness. Material mechanics constraints are embedded in the model's output dynamic threshold. In the time dimension, the dynamic change trend of operating parameters over time can be effectively captured through the recurrent neural network (RNN) or its variant long short-term memory network (LSTM) component. For example, STAN can learn the time-dependent relationship between gearbox speed and load parameters at different times and analyze the delayed response patterns of parameters such as vibration and temperature during speed increase. In the spatial dimension, the attention mechanism can model the spatial distribution relationship of different types of sensor data at the same time. It can automatically assign weights to highlight sensor data features closely related to fault characteristics and weaken irrelevant or interfering data. To improve the robustness of the model, Gaussian noise is injected into the training data to simulate noise environments such as sensor measurement errors and electromagnetic interference that may occur in real applications. During training, the model continuously learns to accurately extract effective features under noise interference, thereby enhancing its resistance to noise and improving its stability in complex real-world environments. For example, the signal-to-noise ratio is set to 20 dB, that is, the ratio of the noise signal power to the original signal power is 1:100; through a large amount of historical test bench data, such as vibration, temperature, speed, and load data, random noise that conforms to the Gaussian distribution is superimposed; The SMOTE algorithm is used to generate virtual fault samples. In gearbox fault data, some fault types may have a small number of samples, resulting in insufficient model learning of these fault characteristics. The SMOTE algorithm generates new virtual fault samples by interpolating in the feature space of minority class fault samples. For example, for a relatively rare fault type like gear tooth breakage, the SMOTE algorithm generates new virtual samples based on the characteristics of existing broken tooth fault samples, increasing the number of broken tooth fault samples. These virtual fault samples are used together with the original fault samples and normal samples for model training, enabling the model to better learn various fault characteristics, especially rare fault characteristics, further improving the model's robustness and fault recognition capabilities. Embed material mechanics constraints into the dynamic threshold generation formula to ensure that the generated threshold has physical meaning and engineering feasibility; For example, for the torque threshold Tth, its upper limit is set to Tth ≤ 0.8σs, where σs is the yield strength of the gearbox material. The yield strength of the material is a key indicator of its mechanical properties. When the torque exceeds a certain percentage of the material's yield strength, the gearbox may undergo plastic deformation or even fail. By incorporating this material mechanics constraint into the threshold generation model, the model automatically considers the material's load-bearing capacity when generating the torque threshold, avoiding the generation of unreasonable thresholds. During the actual training process, the material mechanics constraint is optimized as part of the loss function. When the threshold generated by the model violates the constraint, the loss value is increased, prompting the model to adjust the parameters and generate a dynamic threshold that conforms to the material mechanics properties, thereby improving the reliability and safety of the model in actual engineering applications. Dynamic threshold adjustment module: threshold verification and adjustment, using a reinforcement learning model based on a deep Q network as the discriminant model architecture, and building an intelligent decision-making system to verify and adjust the threshold of a multi-condition gearbox test bench; Record the threshold values ​​over a period of time; These historical thresholds reflect the changing trends of the system's protection thresholds under different operating conditions. For example, under continuous high-speed, high-load conditions, the historical threshold sequence can show how the thresholds are dynamically adjusted as the conditions persist. By analyzing the historical threshold sequence, the DQN can learn the reasonable value range and adjustment mode of the thresholds under similar conditions, providing a reference for validating the current thresholds. Analyze the equipment health index, which combines the health status of the gearbox equipment reflected by vibration sensors, temperature sensors, and torque sensors; For example, an increase in vibration amplitude, an abnormal rise in temperature, or unstable torque fluctuations can all cause the equipment health index to decrease. As part of the state space, the equipment health index enables DQN to take into account the actual operating conditions of the current equipment. A low equipment health index indicates that the gearbox may be on the verge of failure or has already failed. In this case, more careful verification and adjustment of the threshold are required to ensure the effectiveness of the emergency stop protection system. If the system detects five consecutive false alarms, it indicates that the current threshold may be too sensitive and needs to be adjusted. At this time, the threshold increase mechanism is activated with an increase of +Δ, where Δ is between 3% and 10%. The specific increase can be adjusted dynamically according to actual conditions. For example, if false alarms are frequent and severely impact the test progress, a larger Δ value can be selected; if false alarms are relatively minor, a smaller Δ value can be used. In actual operation, the system records relevant information about each false alarm, including the operating parameters, equipment status, and environmental conditions at the time. By analyzing this false alarm data and combining it with the DQN's learning results on the state space, the appropriate threshold increase range is determined. For example, under certain operating conditions, when continuous false alarms occur due to high ambient temperature, analysis shows that the temperature causes deviations in sensor signals, leading to false alarms. In this case, based on the threshold adjustment strategy learned by DQN in high-temperature environments, an appropriate Δ value is selected to increase the threshold to reduce the occurrence of false alarms. When the operating conditions suddenly change, such as a speed step greater than 30% / s, the gearbox's operating state will change dramatically in a short period of time. The original threshold cannot adapt to this sudden change in time, and the dynamic compensation algorithm needs to be triggered. The Kalman filter algorithm is used here to predict threshold drift. Kalman filtering is an efficient recursive filtering algorithm that can optimally estimate the system state based on the system's state equation and observation equation. When the operating conditions suddenly change, the Kalman filter uses existing historical data and current observation data, such as speed, load, vibration, and temperature parameters, to predict the threshold drift. For example, when the speed suddenly rises sharply, the Kalman filter will predict the changes in parameters such as vibration and temperature that may be caused by the sudden change in speed based on the speed change rate and the changing trends of other related parameters, and then predict the direction and magnitude of the threshold drift. Based on the prediction results of the Kalman filter, the system dynamically compensates and adjusts the threshold, allowing the emergency stop protection system to adapt to the sudden change in operating conditions in a timely manner, effectively avoiding the spread of faults caused by threshold lag. During the adjustment process, the Kalman filter will continuously update the prediction results to adapt to the continuous changes in operating conditions, ensuring that the threshold is always within a reasonable and effective range. Control module: Based on the predicted dynamic threshold results, the mechanical braking system is used to achieve braking.

[0014] During the operation of the multi-condition gearbox test bench, the predicted dynamic threshold results were obtained through data processing and analysis, dynamic threshold generation model training, threshold verification and adjustment steps; When the actual operating parameters of the test bench, such as speed, load, vibration, and temperature, are compared with the predicted dynamic thresholds, if the actual parameters are determined to be outside the dynamic threshold range, it indicates that the gearbox is in a faulty or dangerous state. At this time, the central control module will decide to initiate braking measures based on this result; the central control module will use the mechanical braking system to implement braking; The basic principles, main features, and advantages of the present invention are shown and described above. Those skilled in the art should understand that the present invention is not limited to the foregoing embodiments. The foregoing embodiments and descriptions are merely illustrative of the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A multi-operating-condition gearbox test bench emergency stop protection control method, characterized by: include: Step 1: Real-time data collection of the gearbox during operation; Step 2: Align the collected data with timestamps and perform time series feature processing. Based on this time series feature processing, extract the RMS, peak-to-peak value, and kurtosis coefficient of the vibration signal and identify the fault type. Simultaneously, use FFT to analyze the operating status of the gears, and use envelope spectrum to analyze bearing faults. Step 3: Dynamic threshold generation model training. This model uses a spatiotemporal attention network model and is pre-trained using historical testbed data. During adversarial training, Gaussian noise and virtual fault samples are injected to improve the model's robustness. Material mechanics constraints are embedded in the model's output dynamic threshold. Step 4: Threshold verification and adjustment: Using a reinforcement learning model based on a deep Q network as the discriminant model architecture, an intelligent decision-making system is built to verify and adjust the threshold of the multi-condition gearbox test bench. Step 5: Based on the predicted dynamic threshold results, braking is achieved using a mechanical braking system.

2. The emergency stop protection control method for a multi-operating-condition gearbox test bench according to claim 1 is characterized in that: The method for real-time data collection during the operation of the gearbox is as follows: Install speed sensors, load sensors, vibration sensors, and temperature sensors to collect the speed, load, vibration, and temperature parameters of the gearbox in real time during operation.

3. The emergency stop protection control method for a multi-operating-condition gearbox test bench according to claim 1, characterized in that: The specific process of aligning the timestamps of the collected data is as follows: A high-precision clock synchronization module is selected, and clock synchronization equipment based on the Global Positioning System (GPS) is used to provide a unified time reference signal for all sensors. For data with low sampling rates, linear interpolation and spline interpolation methods are used to supplement missing data points on the time axis based on existing data points. For data with high sampling rates, downsampling operations are performed.

4. The emergency stop protection control method for a multi-operating-condition gearbox test bench according to claim 1 is characterized in that: The specific process of extracting the RMS of the vibration signal to identify the fault type is as follows: The square root of the average value of the vibration signal over a period of time is taken and the calculation formula is: , where N is the number of signal sampling points, It is a discrete vibration signal value. When the gearbox operates normally, the vibration RMS value is in a stable range. If the gear is worn or pitting, the vibration energy increases and the RMS value rises accordingly. It can be used as an indicator for early warning of faults.

5. The emergency stop protection control method for a multi-operating-condition gearbox test bench according to claim 1 is characterized in that: The specific process of extracting the peak-to-peak value to identify the fault type is as follows: When a serious fault such as gear tooth breakage occurs, the peak-to-peak value of the vibration increases sharply, exceeding the normal operating range. By continuously monitoring the peak-to-peak value changes, the fault condition can be discovered.

6. The emergency stop protection control method for a multi-operating-condition gearbox test bench according to claim 1 is characterized in that: The specific process of extracting the kurtosis coefficient to identify the fault type is as follows: The kurtosis coefficient is used to measure the impact characteristics of the vibration signal. The calculation formula is: ;in, is the signal mean, is the discrete vibration signal value, and N is the number of signal sampling points. For a normally operating gearbox, the vibration signal kurtosis coefficient fluctuates within a specific range. When a gear has a broken tooth fault, a strong impact is generated, which increases the kurtosis value. This type of fault can be identified based on this feature.

7. The emergency stop protection control method for a multi-operating-condition gearbox test bench according to claim 1 is characterized in that: The specific process of using FFT to analyze the operating status of the gear is as follows: By performing FFT analysis on the vibration signal, the gear meshing frequency and frequency harmonic components can be identified. If the gear is worn, the amplitude at the meshing frequency will change and sideband frequencies will appear. By observing the changes in frequency characteristics, the operating status of the gear can be determined.

8. The emergency stop protection control method for a multi-operating-condition gearbox test bench according to claim 1 is characterized in that: The specific process of analyzing bearing faults using envelope spectrum is as follows: By performing envelope demodulation on the vibration signal and then performing spectrum analysis, the envelope spectrum is obtained; in the envelope spectrum, the fault frequency of the bearing outer ring can be identified. , the calculation formula for the bearing outer ring fault frequency is: ; Where M is the number of bearing rolling elements, is the rotation frequency of the shaft, d is the rolling element diameter, D is the bearing pitch diameter, is the contact angle. Based on the calculated theoretical fault frequency and the envelope spectrum analysis results, it is determined whether the bearing has an outer ring fault.

9. The emergency stop protection control method for a multi-operating-condition gearbox test bench according to claim 1, characterized in that: The specific process of injecting Gaussian noise and virtual fault samples during adversarial training to improve the robustness of the model is as follows: Gaussian noise is injected into the training data to simulate sensor measurement errors and electromagnetic interference noise that occur in actual applications. During the training process, the model learns to extract effective features under noise interference, thereby improving the stability of the model. The SMOTE algorithm generates new virtual fault samples by interpolating in the feature space of minority class fault samples.

10. A multi-operating-condition gearbox test bench emergency stop protection control system, characterized by: include: Data acquisition module: collects data of the gearbox in real time during operation; Data processing module: This module performs time stamp alignment and time series feature processing on the collected data. Based on this processing, it extracts the RMS, peak-to-peak value, and kurtosis coefficient of the vibration signal and identifies the fault type. It also uses FFT to analyze the operating status of the gears and envelope spectrum to analyze bearing faults. Dynamic Threshold Generation Module: This module trains a dynamic threshold generation model using a spatiotemporal attention network model and pre-training with historical test bench data. During adversarial training, Gaussian noise and virtual fault samples are injected to improve the model's robustness. Material mechanics constraints are embedded in the model's output dynamic threshold. Dynamic threshold adjustment module: threshold verification and adjustment, using a reinforcement learning model based on a deep Q network as the discriminant model architecture, and building an intelligent decision-making system to verify and adjust the threshold of a multi-condition gearbox test bench; Control module: Based on the predicted dynamic threshold results, the mechanical braking system is used to achieve braking.

Images