High-horsepower tractor gearbox state on-line monitoring method

Abstract

The invention discloses a high-horsepower tractor gearbox state on-line monitoring method, and relates to the technical field of state monitoring, first, gearbox monitoring data are collected, and the gearbox monitoring data comprise vibration signal data, thermal data, mechanical data, process parameters and oil data; performing multivariate analysis processing on the gearbox monitoring data to obtain first damage data; constructing an improved TimeMixer model, and performing sequence decomposition and mixed path correction on the first damage data to obtain the predicted service life of the gearbox; according to the predicted service life of the gearbox, a tractor gearbox alarm device is arranged, an early warning signal result is generated, graded response from normal monitoring to emergency shutdown is achieved, and a reliable solution is provided for intelligent operation and maintenance of the equipment manufacturing industry.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of state monitoring, in particular to a large-horsepower tractor gearbox state online monitoring method. BACKGROUND

[0002] As the core component of power transmission, the working environment of the tractor gearbox is harsh, the working condition is complex, and new faults are prone to occur. The traditional periodic maintenance or post-maintenance mode based on fixed threshold has the risk of over-maintenance leading to cost increase or sudden failure leading to downtime. Therefore, predictive maintenance of the gearbox has become an inevitable trend.

[0003] At present, an application No. 202310451279.4 discloses a tractor gearbox fatigue life evaluation method based on digital twinning, which comprises: collecting dynamic stress data at the vulnerable position of the tractor gearbox; constructing a virtual three-dimensional model of the tractor gearbox according to the actual use condition, setting stress for the virtual three-dimensional model in the simulation software, and obtaining tractor gearbox fatigue life simulation test data through simulation analysis; fusing the experimental test data of the tractor gearbox fatigue life with the tractor gearbox fatigue life simulation test data, and evaluating the fatigue life of the tractor gearbox through the fused data.

[0004] The above technology focuses on static short sequence analysis and fails to fully explore the degradation law of the gearbox, lacks real-time self-updating prediction ability, and is difficult to self-correct and update according to real-time operation data, which is not sensitive to early gearbox fault features. SUMMARY

[0005] The technical problem solved by the present application is that the degradation law of the gearbox is not fully explored, and the real-time self-updating prediction ability is lacking.

[0006] To solve the above technical problems, the present application provides the following technical scheme: A large-horsepower tractor gearbox state online monitoring method, comprising the following steps: Step S1, collecting gearbox monitoring data, wherein the gearbox monitoring data comprises vibration signal data, thermal data, mechanical data, process parameters and oil data; Step S2, performing multivariate analysis processing on the gearbox monitoring data to obtain first damage data; Step S3, constructing an improved TimeMixer model, performing sequence decomposition and mixed path correction on the first damage data to obtain gearbox predicted life; Step S4, setting a tractor gearbox alarm device according to the gearbox predicted life to generate a warning signal result.

[0007] Preferably, the step S1 comprises the following sub-steps: Step S11, synchronously collecting gearbox monitoring data by using multi-source sensors, the multi-source sensors comprising vibration sensors, temperature sensors, strain sensors and oil quality sensors, marking the positions of the sensors, obtaining gearbox part numbers, deploying vibration sensors on both sides of the input shaft bearing seat, both sides of the output shaft bearing seat and the middle part of the gearbox of the tractor gearbox, and collecting vibration signal data; Step S12, deploying temperature sensors at the outlet of the oil pump, the oil return pipeline, the outer ring of the middle shaft bearing and the clutch, collecting thermal data by using the temperature sensors, deploying strain sensors at the connection part of the bearing seat and the gearbox, collecting mechanical data by using the strain sensors, monitoring the connection part of the bearing seat and the gearbox by using the strain gauge group, obtaining mechanical data, collecting oil data of the lubricating oil circuit by using the oil quality sensor, and collecting process parameters of the tractor according to the CAN bus of the tractor; Step S13, based on a time synchronization protocol, performing time synchronization processing on the vibration signal data, thermal data, mechanical data, process parameters and oil data to generate gearbox monitoring data; The vibration signal data includes three-dimensional vibration acceleration, vibration speed and displacement, the thermal data includes oil temperature, bearing temperature, gearbox surface temperature and environmental temperature, the mechanical data includes input shaft torque, output shaft torque, rotating shaft radial force, axial force and clutch pressure, the process parameters include shaft speed, gear state, throttle opening, load power and working cumulative time, and the oil data includes oil pressure, oil quality and oil pressure.

[0008] Preferably, step S2 further comprises S21, S22 and S23. Performing multivariate analysis processing on the gearbox monitoring data to obtain first damage data; Step S21, setting a fixed length sliding window, performing feature segmentation on the gearbox monitoring data to obtain a plurality of gearbox monitoring data samples, calculating the mean, variance, peak value, kurtosis, RMS value and cumulative time of each gearbox monitoring data sample and saving them as first feature data; Step S22, performing data cross-coupling processing on the first feature data to obtain second feature data; Step S23, performing dimension reduction processing on the second feature data by using the partial least squares method to obtain third feature data; Selecting tractor early operation data for synchronous multivariate analysis processing to obtain health benchmark data, calculating the Mahalanobis distance between the third feature data and the health benchmark based on multivariate statistical theory to obtain first damage data.

[0009] Preferably, the logic of performing data cross-coupling processing on the first feature data is: a temperature-vibration coupling, a load-vibration coupling and a temperature peak coupling are set; The temperature-vibration coupling is used to calculate a ratio of a vibration mean value and a temperature change rate in each time window to obtain a temperature-vibration ratio, the load-vibration coupling is used to calculate a change rate of a vibration speed with an axial force in each time window to obtain a load-vibration coefficient, and the temperature peak coupling is used to calculate a change rate of a bearing temperature gradient and a vibration peak in the time window to obtain a temperature bearing coefficient. The temperature-vibration ratio, the load-vibration coefficient and the temperature bearing coefficient are taken as a coupling feature layer and spliced into the first feature data to obtain second feature data.

[0010] Preferably, the step S3 comprises: An improved TimeMixer model is constructed, a first damage data is set as a sequence D1 with a length L, and the sequence D1 is down-sampled by average pooling to obtain a sequence with a length of L / 2 by averaging every 2 points, a sequence with a length of L / 4 by averaging every 4 points, and a sequence with a length of L / 8 by averaging every 8 points, and so on. For each sequence at a scale, a moving average method is used to extract a trend component of the sequence, an adaptive moving window is set, a window length is dynamically adjusted according to a damage development speed, and a trend component is generated A dominant periodic component of the sequence is extracted based on Fourier transform as a seasonal component ; The seasonal component is corrected by a mixed path The mixed path correction process comprises: Starting from scale 0, the seasonal component is transferred to the seasonal component , and a mixed weighted average is performed to obtain a corrected seasonal component at scale 1, is transferred to scale 2, and a mixed weighted average is performed between to obtain , and the process is stopped when the size M reaches the coarsest size to generate a corrected seasonal component ; The corrected seasonal component is added to the trend component to obtain a second damage sequence; The second damage sequence is taken as an observation value, input to a fully connected layer and predicted to obtain a gearbox predicted life.

[0011] Preferably, the processing logic of the mixed weighted average is: When transferred from scale i to scale i+1, the seasonal component of scale i at the same time point is linearly differentiated

[0012]

[0013]

[0014]

[0015] ​​​​​​​​​​​​​​​​​​The first state is used for displaying the tractor gearbox state as a new machine state, without triggering an alarm signal; The second state is used for displaying the tractor gearbox state as a medium wear state, and triggering a low-frequency alarm signal; The third state is used for displaying the tractor gearbox state as a serious wear state, and triggering a high-frequency emergency alarm signal.

[0016] The present application has the beneficial effects that: through the innovative combination of multi-source data fusion and deep time sequence analysis, the gearbox state life is predicted, and the first damage data accurately reflecting the equipment health state is constructed by using multivariate analysis, thereby overcoming the limitations of a single data source; the TimeMixer deep time sequence model is innovatively used for multi-scale sequence decomposition and mixed path correction, so that long-term damage trends, periodic fluctuations and random noise are effectively separated, a three-level early warning mechanism is established based on the prediction results, and a graded response from normal monitoring to emergency shutdown is realized, which can not only prevent resource waste caused by over-maintenance, but also avoid production interruption caused by sudden failure, thereby providing a reliable solution for the intelligent operation and maintenance of the equipment manufacturing industry. BRIEF DESCRIPTION OF DRAWINGS

[0017] Figure 1 A step flow chart of a large-horsepower tractor gearbox state online monitoring method is provided for an embodiment of the present application. DETAILED DESCRIPTION

[0018] In order to make the above-mentioned objects, features and advantages of the present application more apparent and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments.

[0019] Embodiments, with reference to Figure 1 , a large-horsepower tractor gearbox state online monitoring method is provided, which comprises the following steps: Step S1, collecting gearbox monitoring data, the gearbox monitoring data comprising vibration signal data, thermal data, mechanical data, process parameters and oil data; Step S2, performing multivariate analysis processing on the gearbox monitoring data to obtain first damage data; Step S3, constructing an improved TimeMixer model to perform sequence decomposition and mixed path correction on the first damage data to obtain a gearbox predicted life; Step S4, setting a tractor gearbox alarm device according to the gearbox predicted life to generate an early warning signal result.

[0020] In this embodiment, the gearbox state life is predicted through the innovative combination of multi-source data fusion and deep time series analysis. The first damage data accurately reflecting the health status of the equipment is constructed through multivariate analysis, overcoming the limitations of single data source. The improved TimeMixer deep time series model is innovatively used for multi-scale sequence decomposition and mixed path correction. The main improvements are in four aspects: multi-scale sequence generation, adaptive decomposition method, mixed path correction strategy and prediction framework integration. The multi-scale sequence cooperates with the fine scale and coarse scale sequence, providing early abnormal detection sensitivity, ensuring long-term trend prediction stability, reducing false negatives and false positives between scales, and effectively reducing the mean absolute error compared with the traditional TimeMixer model. The early fault detection rate and prediction length are improved. In addition, the dynamic window mechanism can adaptively learn the damage evolution law of the specific device, and the multi-scale average pooling can effectively suppress high-frequency noise. The mixed path correction automatically compensates for the lack of information, providing a generalizable technical framework for intelligent operation and maintenance of rotating machinery equipment in tractors. The intelligent early warning device based on the prediction results establishes a three-level early warning mechanism, realizes graded response from normal monitoring to emergency shutdown, prevents resource waste caused by over-maintenance, avoids production interruption caused by sudden failure, and provides a reliable solution for intelligent operation and maintenance of equipment manufacturing industry.

[0021] Step S1 includes the following sub-steps: Step S11, synchronously collecting gearbox monitoring data by using multi-source sensors, the multi-source sensors including vibration sensors, temperature sensors, strain sensors and oil quality sensors, marking and numbering the positions where the sensors are deployed to obtain the gearbox part numbers, deploying vibration sensors on both sides of the input shaft bearing seat, both sides of the output shaft bearing seat and the middle part of the gearbox of the tractor, and collecting vibration signal data; Step S12, deploying temperature sensors at the outlet of the oil pump, the oil return pipeline, the outer ring of the shaft bearing and the clutch, collecting thermal data by using the temperature sensors, deploying strain sensors at the connection part of the bearing seat and the gearbox, collecting mechanical data by using the strain sensors, monitoring the connection part of the bearing seat and the gearbox by using the strain gauge group to obtain mechanical data, collecting oil data of the lubricating oil circuit by using the oil quality sensor, and collecting process parameters of the tractor according to the CAN bus of the tractor; Step S13, based on the time synchronization protocol, performing time synchronization processing on the vibration signal data, thermal data, mechanical data, process parameters and oil data to generate gearbox monitoring data; The vibration signal data includes three-dimensional vibration acceleration, vibration speed and displacement, the thermal data includes oil temperature, bearing temperature, box surface temperature and environment temperature, the mechanical data includes input shaft torque, output shaft torque, rotating shaft radial force, axial force and clutch pressure, the process parameters include shaft rotating speed, gear state, throttle opening, load power and working cumulative length, and the oil data includes oil pressure, oil quality and oil pressure.

[0022] One three-axis vibration sensor is arranged on each side of the gearbox input shaft bearing seat and fixedly installed by using M5 thread, and one three-axis vibration sensor is arranged on each side of the output shaft bearing seat and installed in the same direction as the input shaft sensor. One single-axis vibration sensor is arranged at the geometric center of the box body for monitoring the overall vibration level. All vibration sensors use a sampling frequency of 25.6 kHz to ensure the capture of high-frequency components such as gear meshing frequency. A PT100 platinum resistance temperature sensor is installed on the oil pump outlet pipeline and threadedly installed on the oil way tee joint. The same type of temperature sensor is installed on the oil return pipeline to monitor the heat exchange efficiency of the lubricating oil. A thermocouple sensor is embedded in the box near the intermediate shaft bearing outer ring and uses heat-conducting glue to ensure good thermal contact. An infrared temperature sensor is installed outside the clutch housing to non-contact monitor the temperature of the friction plate. Strain gauges are pasted at the connection part between the bearing seat and the box body, arranged in 90° or 45°, fixed by using special adhesive, and covered with a protective coating to prevent damage. The strain gauges are connected by a Wheatstone bridge circuit to monitor the dynamic stress changes in real time. An online oil quality sensor is installed in series in the main oil circulation system to monitor parameters such as lubricating oil viscosity, moisture content and particle contamination. All data acquisition devices are operated synchronously by using the clock synchronization protocol PTP, and the data with different sampling frequencies are resampled. In this embodiment, through the targeted arrangement of key parts, all important components of the gearbox are ensured to be within the effective monitoring range, avoiding the problem of missed detection due to monitoring blind area.

[0023] Step S2 further includes S21, S22 and S23. The gearbox monitoring data is subjected to multivariate analysis processing to obtain first damage data. In step S21, a fixed length sliding window is set, the gearbox monitoring data is subjected to feature segmentation, a plurality of gearbox monitoring data samples are obtained, the mean, variance, peak value, kurtosis, RMS value and cumulative length of each gearbox monitoring data sample are calculated and saved as first feature data. In step S22, the first feature data is subjected to data cross-coupling processing to obtain second feature data. In step S23, the second feature data is subjected to dimension reduction processing by using the partial least squares method to obtain third feature data. The early operation data of the tractor is selected for multivariate analysis processing to obtain health benchmark data, and Mahalanobis distance between the third characteristic data and the health benchmark is calculated based on multivariate statistical theory to obtain first damage data.

[0024] The arithmetic mean, variance, maximum absolute value in the window, root mean square value, and actual time length of valid data in the window are calculated for all data points in each window to establish a feature database, and each feature sample contains a timestamp, a sensor ID, and six characteristic values, wherein the feature database is stored according to the timestamp and sensor position number index to generate first characteristic data; A dynamic correlation matrix of vibration features and temperature features is established to analyze the sensitivity difference of various features to load changes, and product coupling, ratio coupling, and difference coupling are adopted to obtain second characteristic data, and PLS is used to reduce the dimension of the second characteristic data to obtain third characteristic data with key features. The data of 500-1000 hours at the beginning of the operation of the gearbox is selected as the monitoring benchmark to ensure that the monitoring data covers typical working conditions (empty load, light load, and heavy load), and the health state data is processed in the same way as S21-S23 to obtain feature data under the health state, and the mean and covariance matrix thereof are calculated as health benchmark data. The Mahalanobis distance between the health benchmark data and the third characteristic data is calculated to generate first damage data, wherein the Mahalanobis distance considers the correlation between features, which is better than the Euclidean distance and is beneficial to the multivariate evaluation of the gearbox damage. In this embodiment, the precise conversion from the original monitoring data to the comprehensive damage index is realized, and a scientific and reliable technical foundation is provided for the health state evaluation of the gearbox.

[0025] The logic of cross-coupling processing of the first characteristic data is as follows: Temperature-vibration coupling, load-vibration coupling, and temperature peak coupling are set. The temperature-vibration coupling is used to calculate the ratio of vibration mean value and temperature change rate in each time window to obtain a temperature-vibration ratio, the load-vibration coupling is used to calculate the change rate of vibration speed with axial force in each time window to obtain a load-vibration coefficient, and the temperature peak coupling is used to calculate the change rate of bearing temperature gradient and vibration peak value in the time window to obtain a temperature bearing coefficient. The temperature-vibration ratio, load-vibration coefficient, and temperature bearing coefficient are taken as coupling feature layers and spliced into the first characteristic data to obtain second characteristic data.

[0026] Temperature-vibration coupling (temperature-vibration ratio): For each time window, the mean of the vibration signal and the rate of temperature change are calculated. The rate of temperature change can be calculated by subtracting the temperature value of the previous time window from the temperature value of the current time window and then dividing by the duration of the time window. Load-vibration coupling (load-vibration coefficient): The rate of change of vibration velocity with axial force within each time window is calculated. The axial force comes from the mechanical data measured by the strain sensor. Temperature peak coupling (temperature-bearing coefficient): The rate of change of bearing temperature gradient and vibration peak value within the time window is calculated. The bearing temperature gradient is the rate of change of bearing temperature, and the vibration peak value is the peak value of vibration within each window. Temperature-bearing coefficient = bearing temperature change rate / vibration peak value change rate. In this embodiment, by coupling multiple physical quantities, the limitations of single sensor information can be overcome, providing more comprehensive equipment status information for subsequent processing. Coupling features can amplify small changes caused by faults, improving the detection capability of early faults. By fusing information from multiple sensors, the false alarm rate can be reduced. Each coupling feature has a clear physical meaning, which is convenient for interpretation and diagnosis.

[0027] Step S3 includes: An improved TimeMixer model is constructed. The first damage data is set as a sequence D1 of length L. Through average pooling downsampling, the sequence D1 is averaged at every 2 points to obtain a sequence of length L / 2. Then, the sequence D1 is averaged at every 4 points to obtain a sequence of length L / 4. This process is repeated to obtain sequences of M scales. For each scale of the sequence, the moving average method is used to extract the trend components. An adaptive moving window is set, and the window length is dynamically adjusted according to the damage development speed to generate the trend components. Based on Fourier transform, the dominant periodic components of the sequence are extracted as seasonal components. ; Correcting seasonal components using mixed pathways The hybrid path correction process includes: Starting from scale 0, seasonal components Transmission to seasonal components Then, a mixed weighted average is performed to obtain the seasonal components at scale 1 after correction. ,Will Transmitted to scale 2, and with Perform a mixed weighted average to obtain The process continues until dimension M reaches its coarsest point, at which point the corrected seasonal components are generated. ; The revised seasonal components With trend components Add them together to obtain the second damage sequence; The second damage sequence is input into a fully connected layer as an observation value and prediction is performed to obtain the gearbox predicted life.

[0028] In this embodiment, the first damage data is sequenced to set the first damage data as a time sequence with a length of L, each point representing the damage degree of a time window, and then down-sampling is performed, wherein the down-sampling quality ensures that the length of each scale sequence is at least 10, and for cases that cannot be evenly divided, a truncation strategy is adopted, and the actual time resolution of each scale is recorded. Down-sampling can reduce data processing and improve calculation speed. Then, the mixed path correction seasonal component is performed. Starting from the finest scale (initial sequence), the seasonal component is transmitted to the next coarser scale (scale 1), and a mixed weighted average is performed with the original seasonal component of scale 1 to obtain the corrected seasonal component of scale 1. The process is repeated until the coarsest scale M. The corrected seasonal component of each scale is added to the corresponding trend component to obtain the reconstructed sequence of each scale. Then, the reconstructed sequences of multiple scales are up-sampled to the original length (by interpolation), and the sequences are fused (weighted average) to obtain the second damage sequence. The second damage sequence is input into a fully connected layer as an observation value and prediction is performed. The fully connected layer can learn the mapping from the second damage sequence to the remaining life; The method decomposes the time sequence into trend and seasonal components through multi-scale analysis, and corrects the seasonal components at multiple scales to better capture the long-term trend and periodic patterns of the time sequence. The mixed path correction preserves the information flow between different scales, so that the seasonal component of the coarse scale can benefit from the details of the fine scale, and the seasonal component of the fine scale can also be constrained by the stability of the coarse scale. Finally, the corrected sequence is predicted by the fully connected layer to improve the accuracy of life prediction.

[0029] The processing logic of the mixed weighted average is as follows: When transmitting from scale i to scale i+1, the seasonal component of scale i at the same time point is linearly interpolated The interpolated values are added and averaged to generate the seasonal component point mean The seasonal component point mean is weighted and averaged with the seasonal component The data in the same time scale is weighted and averaged to calculate the weighted average, and the calculation expression of the weighted average is: ; Wherein, is the corrected seasonal component of scale i, is the seasonal component point mean of scale i.

[0030] In this embodiment, the seasonal component S_i at scale i is upsampled (linearly interpolated) to make its length the same as the sequence length at scale i+1, but its time point is aligned to the time point at scale i+1. In fact, since scale i+1 is a downsampled version of scale i, the sequence at scale i needs to be interpolated onto the time grid of scale i+1. Then, the seasonal component at scale i (denoted as S_i) is... The seasonal component S{i+1} at scale i+1 is weighted and averaged to obtain the corrected seasonal component S{i+1}' at scale i+1. Through this mixed weighted average, the seasonal component at the fine scale contains more detailed periodic fluctuation information, while the seasonal component at the coarse scale reflects the overall periodic pattern. By transferring the information from the fine scale to the coarse scale, the seasonal component at the coarse scale can be corrected to retain more detailed periodic characteristics. At the same time, the weighted average can balance the details of the fine scale and the stability of the coarse scale, avoiding overfitting of the noise at the fine scale.

[0031] The processing logic for using the second damage sequence as an observation, inputting it into the fully connected layer, and performing prediction is as follows: A multilayer perceptron is constructed with two hidden layers. The dimension of the second damage sequence is set to d. The second damage sequence is input into the first hidden layer for processing and ReLU activation, outputting the first layer result. The number of neurons in the first hidden layer is d / 2. The first layer result is input into the second hidden layer for processing and ReLU activation, generating the second layer result. The number of neurons in the second hidden layer is d / 4. The second layer result is input into the output layer, with the number of neurons set to 1. The output is restricted to between 0 and 1 using the Sigmoid activation function, generating the gearbox prediction lifespan.

[0032] In this embodiment, the second damage sequence is used as the input feature, and the normalized remaining lifespan is used as the target variable. The dimensionality is gradually reduced through two hidden layers to extract key features. At the same time, the ReLU activation function is used to introduce nonlinearity, enabling the model to learn complex nonlinear relationships. The output layer uses the Sigmoid function to limit the output to between 0 and 1, representing the percentage of the device's remaining lifespan. Then, it can be converted into a specific time unit according to the actual maximum lifespan. This embodiment avoids overfitting by gradually reducing dimensionality and extracting the most important key features.

[0033] Step S4 specifically includes: Based on the maximum lifespan of the transmission, calculate the predicted lifespan in days. Predicted lifespan in days = Predicted lifespan of the transmission. The maximum lifespan of the transmission is determined by setting up a tractor transmission alarm device to provide real-time warnings of transmission wear signals. The transmission lifespan status is determined based on the predicted lifespan days of the transmission, triggering the generation of warning signals.

[0034] In this embodiment, the damage is quantified as the tractor gearbox life in days by the early warning device, the prediction result is converted into early warning action, and factors such as damage degree trend are considered to ensure the accuracy and timeliness of the early warning. At the same time, the system has a feedback mechanism, which can adjust the early warning state according to the confirmation of personnel and provide decision support to guide maintenance work. Therefore, the early warning device of the tractor gearbox can realize intelligent state monitoring and early warning, and improve the reliability and safety of the equipment.

[0035] The specific method for determining the gearbox life state according to the gearbox predicted life in days is as follows: The determination threshold is set as a first threshold, a second threshold and a third threshold. When the gearbox predicted life in days is in the first threshold, the gearbox is in a first state, and the early warning signal result is normal. When the gearbox predicted life in days is in the second threshold, the gearbox is in a second state, and the early warning signal result is attention. When the gearbox predicted life in days is in the third threshold, the gearbox is in a third state, and the early warning signal result is warning. The first threshold is greater than the second threshold, which is greater than the third threshold.

[0036] In this embodiment, the normal life of the gearbox is set to 1000 days, the first threshold is set to more than 60 days, the second threshold is set to 30-60 days, and the third threshold is set to 15-30 days. The early warning signal result can change with the color of the indicator light, the warning information on the display screen and the sound of the buzzer.

[0037] The first state, the second state and the third state include: The first state is used to display the tractor gearbox state as a new machine state without triggering an alarm signal. The second state is used to display the tractor gearbox state as a medium wear state and trigger a low-frequency alarm signal. The third state is used to display the tractor gearbox state as a serious wear state and trigger a high-frequency emergency alarm signal.

[0038] The first state is set to not trigger an alarm, or a normal state signal (such as a green indicator light) can be triggered. The second state is set to trigger a low-frequency alarm signal, such as a yellow light that flashes once a minute, or a sound prompt that is emitted every certain period of time. The third state is set to trigger a high-frequency emergency alarm signal, such as a red light that flashes once a second, a continuous sound alarm, and the like. By monitoring the damage data, the health status of the gearbox can be assessed in real time, and different levels of alarms can be provided according to the severity of the state, so as to remind the operator to take appropriate measures.

[0039] Those skilled in the art will appreciate that embodiments of the present application can be readily used as a method, a system or a computer program product. Accordingly, the present application can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application can take the form of a computer program product on one or more computer-usable storage media (or computer- readable storage media) having computer-usable program code embodied in the medium. The medium can be any available medium or combination thereof that is accessible by a general purpose or special purpose computer. By way of example, such computer-usable storage media can include a volatile memory, such as a random access memory (RAM), a non-volatile memory, such as a read-only memory (ROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic disk, a flash memory, a compact disk (CD) or a digital versatile disk (DVD). The computer-usable program code can include any suitable set of instructions, statements or Figure 1 one or more functions specified in the flow or flows and / or blocks Figure 1 one or more functions specified in the flow or flows and / or blocks

[0040] It should be noted that the above-mentioned embodiments are only used to illustrate but not to limit the technical solutions of the present application. Although the present application is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced equivalently without departing from the spirit and scope of the technical solutions of the present application, and they should be covered in the scope of the claims of the present application.

Claims

1. A method for online monitoring of the gearbox status of a high-horsepower tractor, characterized in that, Includes the following steps: Step S1: Collect transmission monitoring data, which includes vibration signal data, thermal data, mechanical data, process parameters, and oil data; Step S2: Perform multivariate analysis on the transmission monitoring data to obtain the first damage data; Step S3: Construct an improved TimeMixer model to perform sequence decomposition and hybrid path correction on the first damage data to obtain the predicted life of the gearbox. Step S4: Based on the predicted lifespan of the gearbox, set the tractor gearbox alarm device and generate a warning signal result.

2. The method for online monitoring of the gearbox status of a high-horsepower tractor as described in claim 1, characterized in that, Step S1 includes the following sub-steps: Step S11: Synchronously collect transmission monitoring data using multi-source sensors, including vibration sensors, temperature sensors, strain sensors, and oil quality sensors. Mark and number the locations where the sensors are deployed to obtain the transmission part number. Deploy vibration sensors on both sides of the input shaft bearing seat, both sides of the output shaft bearing seat, and the middle of the gearbox of the tractor transmission to collect vibration signal data. Step S12: Temperature sensors are deployed at the oil pump outlet, return oil line, outer ring of the central shaft bearing, and clutch to collect thermal data. Strain sensors are deployed at the connection between the bearing housing and the housing to collect mechanical data. Strain gauges are used to monitor the connection between the bearing housing and the housing to obtain mechanical data. Oil quality sensors are used to collect oil data in the lubricating oil circuit. Tractor process parameters are collected according to the tractor CAN bus. Step S13: Based on the time synchronization protocol, the vibration signal data, thermal data, mechanical data, process parameters and oil data are processed for time synchronization to generate transmission monitoring data; The vibration signal data includes three-dimensional vibration acceleration, vibration velocity, and displacement; the thermodynamic data includes oil temperature, bearing temperature, housing surface temperature, and ambient temperature; the mechanical data includes input shaft torque, output shaft torque, radial force of rotating shaft, axial force, and clutch pressure; the process parameters include shaft speed, gear status, throttle opening, load power, and cumulative working time; and the hydraulic data includes oil pressure, oil quality, and oil pressure.

3. The method for online monitoring of the gearbox status of a high-horsepower tractor as described in claim 2, characterized in that, Step S2 also includes S21, S22, and S23; The transmission monitoring data is processed by multivariate analysis to obtain the first damage data; Step S21: Set a fixed-length sliding window, perform feature segmentation on the transmission monitoring data, obtain several transmission monitoring data samples, calculate the mean, variance, peak value, kurtosis, RMS value and cumulative duration of each transmission monitoring data sample and save them as the first feature data; Step S22: Perform data cross-coupling processing on the first feature data to obtain the second feature data; Step S23: Use partial least squares to perform dimensionality reduction on the second feature data to obtain the third feature data; The early operating data of the tractor were selected and multivariate analysis was performed simultaneously to obtain health baseline data. Based on multivariate statistical theory, the Mahalanobis distance between the third characteristic data and the health baseline was calculated to obtain the first damage data.

4. The method for online monitoring of the gearbox status of a high-horsepower tractor as described in claim 3, characterized in that, The logic for performing data cross-coupling processing on the first feature data is as follows: Configure temperature vibration coupling, load vibration coupling, and temperature peak coupling; The temperature-vibration coupling is used to calculate the ratio of the vibration mean to the temperature change rate in each time window to obtain the temperature-vibration ratio. The load-vibration coupling is used to calculate the rate of change of vibration velocity with axial force in each time window to obtain the load-vibration coefficient. The temperature peak coupling is used to calculate the rate of change of bearing temperature gradient and vibration peak in the time window to obtain the temperature-bearing coefficient. The temperature-vibration ratio, load vibration coefficient, and temperature bearing coefficient are used as coupling feature layers and spliced ​​into the first feature data to obtain the second feature data.

5. The method for online monitoring of the gearbox status of a high-horsepower tractor as described in claim 4, characterized in that, Step S3 includes: Construct an improved TimeMixer model, the improved TimeMixer model comprising: The first damage data is set as a sequence D1 of length L. By downsampling with average pooling, the sequence D1 is averaged at every 2 points to obtain a sequence of length L / 2. Then, the sequence D1 is averaged at every 4 points to obtain a sequence of length L / 4. This process is repeated to obtain sequences of M scales. For each scale of the sequence, the moving average method is used to extract the trend components. An adaptive moving window is set, and the window length is dynamically adjusted according to the damage development speed to generate the trend components. The dominant periodic component of the sequence is extracted based on Fourier transform as the seasonal component. ; Correcting the seasonal components using a mixed pathway The hybrid path correction process includes: Starting from scale 0, seasonal components Transmission to seasonal components Then, a mixed weighted average is performed to obtain the seasonal components at scale 1 after correction. ,Will Transmitted to scale 2, and with Perform a mixed weighted average to obtain The process continues until dimension M reaches its coarsest point, at which point the modified seasonal component is generated. ; The revised seasonal components With trend components Add them together to obtain the second damage sequence; The second damage sequence is used as an observation, input into the fully connected layer, and predicted to obtain the gearbox's predicted lifespan.

6. The method for online monitoring of the gearbox status of a high-horsepower tractor as described in claim 5, characterized in that, The processing logic for the hybrid weighted average is as follows: When transferring from scale i to scale i+1, the seasonal component of scale i at the same time point is obtained by using the linear interpolation method. The average of the sums is calculated to generate the mean of the seasonal component points. The mean of the seasonal component points and the seasonal component The data within the same time scale are weighted and averaged. The expression for the weighted average is as follows: ； in, The seasonal component after scale i correction. Let be the mean of the seasonal component points at scale i.

7. The method for online monitoring of the gearbox status of a high-horsepower tractor as described in claim 6, characterized in that, The processing logic for using the second damage sequence as an observation, inputting it into the fully connected layer, and performing prediction is as follows: A multilayer perceptron is constructed with two hidden layers. The dimension of the second damage sequence is set to d. The second damage sequence is input into the first hidden layer for processing and ReLU activation, outputting the first layer result. The number of neurons in the first hidden layer is d / 2. The first layer result is input into the second hidden layer for processing and ReLU activation, generating the second layer result. The number of neurons in the second hidden layer is d / 4. The second layer result is input into the output layer, with the number of neurons set to 1. The output is restricted to between 0 and 1 using the Sigmoid activation function, generating the gearbox prediction lifespan.

8. The method for online monitoring of the gearbox status of a high-horsepower tractor as described in claim 7, characterized in that, Step S4 specifically includes: Based on the maximum lifespan of the transmission, the predicted lifespan in days is calculated. The expression for calculating the predicted lifespan in days is as follows: Transmission predicted life days = Transmission predicted life The maximum lifespan of the transmission is determined by setting up a tractor transmission alarm device to provide real-time warnings of transmission wear signals. The transmission lifespan status is determined based on the predicted lifespan days of the transmission, triggering the generation of warning signals.

9. The method for online monitoring of the gearbox status of a high-horsepower tractor as described in claim 8, characterized in that, The specific method for determining the lifespan status of a transmission based on its predicted lifespan in days is as follows: The judgment thresholds are set as the first threshold, the second threshold, and the third threshold. When the predicted lifespan days of the transmission are at the first threshold, the transmission is in the first state and the warning signal result is normal. When the predicted lifespan days of the transmission are at the second threshold, the transmission is in the second state, and the warning signal result is "attention". When the predicted lifespan days of the transmission are at the third threshold, the transmission is in the third state, and the warning signal result is a warning. The first threshold is greater than the second threshold, which is greater than the third threshold.

10. The method for online monitoring of the gearbox status of a high-horsepower tractor as described in claim 9, characterized in that, The first state, the second state, and the third state include: The first state is used to display that the tractor gearbox is in a new machine state, and no alarm signal needs to be triggered; The second state is used to indicate that the tractor gearbox is in a state of moderate wear and to trigger a low-frequency alarm signal; The third state is used to indicate that the tractor gearbox is in a severely worn state and to trigger a high-frequency emergency alarm signal.

Images