Agricultural tractor intelligent fault diagnosis system and method based on Internet of Things

Abstract

The invention belongs to the technical field of agricultural tractors, and particularly relates to an agricultural tractor intelligent fault diagnosis system and method based on the Internet of Things, and the system comprises a sensor assembly, a data collection module, a wireless transmission module, a cloud server, and a user terminal. The components are combined to form a component installation stage, a preliminary debugging stage and a comprehensive debugging stage, and system operation is carried out through the three stages. The sensor assembly comprises a vibration sensor, a temperature sensor and a smoke sensor. According to the method, fault detection and repair are carried out in each key stage of the assembly process, so that the assembly quality of each part is ensured; service is provided for final vehicle off-line fault diagnosis, and it is ensured that detection is more efficient and accurate after vehicle assembly is completed; full-cycle fault diagnosis is realized, and the precision and efficiency of fault detection are improved; the rework rate and the maintenance cost are reduced, and the reliability and the service life of the whole vehicle are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of agricultural tractors, and mainly relates to an intelligent fault diagnosis system and method for agricultural tractors based on the Internet of Things. Background Art

[0002] As an important device in modern agricultural production, agricultural tractors have gradually replaced traditional manual tools and become an important symbol of agricultural modernization. However, there is still room for improvement in the technology and reliability of agricultural tractors. In the actual production process, problems with the assembly quality of tractors may lead to various forms of failures, affecting normal operations. Drivers and maintenance personnel need to clarify the common fault causes of agricultural tractors and their solutions in order to promptly eliminate faults, reduce repairs caused by assembly quality problems, and ensure the operation progress. The use of agricultural tractors is highly seasonal. Especially during the busy farming season, it often leads to downtime for repairs due to assembly quality problems, missing the best operation time and seriously affecting the economic interests of farmers. Therefore, the assembly quality inspection of tractors is crucial for the development of agricultural modernization and intelligence and is an important problem that must be solved. Full-cycle fault diagnosis can detect and repair problems in the early stage of the assembly process, avoiding situations where problems accumulate and are difficult to detect and repair after the completion of the whole machine assembly. Timely detection and repair of faults avoid rework after the completion of the whole machine assembly and reduce maintenance costs. Stage-by-stage inspection ensures the gradual improvement of assembly quality, avoids large-scale rework, and improves production efficiency.

[0003] In the existing field of tractor assembly inspection, there is a lack of effective assembly quality inspection equipment and methods, especially insufficient means for comprehensively diagnosing the assembly situation of the whole tractor. This results in the fact that after the tractor comes off the production line, it is prone to failures due to only single inspection, affecting the product's reputation and sales and reducing the user's operation efficiency. For this reason, an intelligent fault diagnosis system for agricultural tractors based on the Internet of Things is proposed, which can monitor and diagnose the states of key components in real time during the assembly process. Through multi-sensor data fusion and big data analysis, a comprehensive evaluation of the assembly quality of the whole machine is realized, improving the efficiency and accuracy of fault diagnosis, and ultimately improving the reliability and service life of the equipment. The Internet of Things technology enables fault diagnosis to be carried out in real time at each stage of the assembly process, ensuring that problems are discovered in a timely manner. By using cloud servers and big data analysis technology, a large amount of sensor data can be efficiently processed and analyzed, improving the accuracy of fault detection. Through the Internet of Things technology, users can remotely monitor the equipment status, receive fault warnings and maintenance suggestions in a timely manner, improving the reliability and service life of the equipment. Based on big data analysis and machine learning algorithms, it is possible to predict the equipment status and give early warnings of potential faults, notifying users in advance for maintenance to avoid the expansion of faults. Summary of the Invention

[0004] To address the deficiencies of the above-mentioned existing technologies, the objective of the present invention is to provide an intelligent fault diagnosis system and method for agricultural tractors based on the Internet of Things. This system and method can achieve fault diagnosis throughout the entire assembly cycle, ensuring that problems are promptly detected and repaired during the assembly process, avoiding the situation where problems accumulate and are difficult to detect and repair after the complete assembly of the whole machine, thereby reducing maintenance costs and improving production efficiency and equipment reliability.

[0005] To achieve the above objective, the present invention adopts the following technical solutions:

[0006] An intelligent fault diagnosis system for agricultural tractors based on the Internet of Things includes a sensor assembly, a data acquisition module, a wireless transmission module, a cloud server, and a user terminal. When combined, the above components jointly form a component installation stage, a preliminary debugging stage, and a comprehensive debugging stage, and the system operates through these three stages. The sensor assembly includes a vibration sensor, a temperature sensor, and a smoke sensor. Among them, the vibration sensor uses an ADXL345 three-axis accelerometer, the temperature sensor selects an S18B20 digital temperature sensor, and the smoke sensor selects an ATRS-1031. Each sensor is mainly installed on the crankcase, cylinder head, bearing seat, input shaft, output shaft, gearbox, etc., and corresponding sensors should also be installed for detection when other parts are off the production line. The data acquisition module includes an engine acquisition module and a gearbox acquisition module, which are respectively installed near the engine, near the gearbox, and on the brackets of other parts. The wireless transmission module is a LoRa module. The cloud server is linked to the whole machine detection platform, processes and analyzes the data through big data analysis and machine learning algorithms, generates fault warning information, and sends it to the user terminal.

[0007] The vibration sensor using an ADXL345 three-axis accelerometer is respectively installed at the central position on the outer wall of the crankcase, the middle position on the top of the cylinder head, directly above the bearing seat, on the outer shell near the input shaft of the gearbox, on the outer shell near the output shaft of the gearbox, and on the outer shell of the gearbox. The temperature sensor selects an S18B20 digital temperature sensor and is respectively installed on the side of the cylinder head and the outer shell of the gearbox.

[0008] The data acquisition module includes a microcontroller and a data storage unit, which are respectively installed on the brackets near the engine and on the brackets near the gearbox.

[0009] The microcontroller uses an ESP32 microcontroller, which has powerful processing capabilities and wireless communication functions. The data storage unit uses an SD card module to temporarily store sensor data to prevent data loss. The wireless transmission module uses a LoRa module to achieve long-distance data transmission and ensure that data can be promptly transmitted to the cloud server.

[0010] The cloud server uses Google Cloud for data storage and management, and utilizes its built-in distributed database system to process big data. The following method is used for big data processing and calculation:

[0011] Step A: Construct a time-domain signal sequence X n (t) with noise, and construct a white noise time-domain signal sequence ω n (t) with a standard deviation of ε0. Combine the white noise signal with the original signal X n (t) to construct a new signal as follows: X n (t) = X(t) + ε0ω n (t), where n = 1, 2,..., K

[0012] Step B: Use EMD to decompose the constructed time-domain signal sequence X n (t) with noise, extract the first-order IMF component, and calculate the average value of its signal amplitude.

[0013]

[0014] After the first-level IMF decomposition, the residual of the signal is:

[0015]

[0016] Step C: Based on the residual, according to the signal processing method in Step S1, reconstruct the signal r(t) + ε0EMD1(ω n (t)) by adding white noise. Extract the second-order IMF component through EMD decomposition. The calculation methods for the average value of the amplitude of this component signal and the new residual are shown in the following formula:

[0017]

[0018] Step D: Repeat the signal decomposition according to the same rule as shown in the following formula until calculating the residual r k (t) of the k-th order IMF and the average value of the (k + 1)-th order IMF until the residual r k (t) satisfies the condition of Equation , or the signal cannot be further decomposed by EMD.

[0019]

[0020] In the formula, T is the length of the original signal X(t); r k (t) is the residual signal of the k-th order IMF.

[0021] In the component installation stage, vibration sensors and temperature sensors are installed at key positions of the engine and the transmission; in the preliminary commissioning stage, the positions of the sensors in the component installation stage are kept unchanged, and the installed sensors are continued to be used; in the comprehensive commissioning stage, the operating state of the whole machine is comprehensively detected to ensure the normal collaborative work of all components, and potential faults are identified and repaired.

[0022] The present invention also provides an intelligent fault diagnosis method for an agricultural tractor based on the Internet of Things using the intelligent fault diagnosis system as described above, including the following steps:

[0023] Step S1: Start the simulation operation platform. During the assembly process, the input shafts of the engine and the transmission are driven by a motor to simulate their actual operating states. The simulation operation platform can provide a stable power source to enable each component to operate in a simulation state. The purpose of the simulation operation is to ensure the performance and state of each component during individual operation and provide basic data for subsequent detection.

[0024] Step S2: Data acquisition. The sensors collect vibration and temperature signals generated during the simulation operation in real time. The vibration sensors monitor the vibration conditions of each key position, including frequency, amplitude, and transient events. The temperature sensors monitor the temperature changes of each key position, including the temperature rise rate and the maximum temperature. The collected data is preliminarily processed by the acquisition module through the CEEMDAN algorithm and transmitted to the cloud server through the LoRa wireless transmission module.

[0025] Step S3: Data preprocessing. After receiving the data, the cloud server first performs data preprocessing. The Pandas library is used to clean the data, removing outliers and missing values. The MinMaxScaler is used to normalize the data to the range of 0-1 to ensure that different features have the same magnitude. The Butterworth low-pass filter is used to filter the noise of the data, removing high-frequency noise and retaining the main components of the signal. The preprocessed vibration and temperature signals are decomposed by CEEMDAN to obtain a number of intrinsic mode functions (IMFs), and each IMF represents the oscillatory component of the signal at different time scales.

[0026] Step S4: Feature extraction. Wavelet transform is performed on the IMFs obtained by CEEMDAN to extract the time-frequency features of the signal. Wavelet transform is a time-frequency analysis method that can effectively detect transient events and local features in the signal. By performing wavelet transform on the IMFs obtained by CEEMDAN decomposition, local features of the signal at different scales and positions can be obtained. Commonly used wavelet functions include Morlet wavelet, Mexican Hat wavelet, and Ricker wavelet.

[0027] Step S5: Pattern recognition and fault diagnosis. Pattern recognition and fault diagnosis using Heterogeneous Oblique Random Forest can be carried out according to the following steps:

[0028] Data preparation: Collect and preprocess data. As described in the previous step, this includes feature selection and data normalization.

[0029] Model training: Use the training data to build a Heterogeneous Oblique Random Forest model. Each decision tree node uses an oblique hyperplane to divide the data to improve the classification ability of the model.

[0030] Model evaluation: Evaluate the model performance through methods such as cross-validation, and adjust the parameters to optimize the accuracy.

[0031] Pattern recognition and fault diagnosis: Apply the model to predict the class or state of new data, identify specific patterns or detect potential faults.

[0032] Result analysis: Analyze the model output, determine the key features and patterns, and diagnose the sources of potential faults.

[0033] Step S6: Fault judgment and warning. According to the preset fault detection criteria, judge whether the vibration signal and temperature signal are abnormal. The detection criteria for the vibration signal include the enhancement of frequency components, the peak value of the vibration amplitude, etc. The detection criteria for the temperature signal include the rate of temperature rise and the maximum temperature. When an abnormal signal is detected, the cloud server generates a fault warning message and sends a warning notification through the user terminal. The user terminal can monitor the device status in real time, receive the fault warning message in a timely manner, and provide maintenance suggestions. Specifically, the present invention performs detections at each key stage of the assembly process to ensure the gradual improvement of the assembly quality and to timely discover and repair problems that occur during the assembly process.

[0034] In step S4, the Morlet wavelet is used as the mother wavelet function for feature extraction. The specific steps are as follows:

[0035] Step ①: Select the wavelet function and scale. Select the Morlet wavelet as the mother wavelet function and determine the scale range of the wavelet transform. In the present invention, 1 to 31 is selected as the scale range to capture the characteristics of different frequency components.

[0036] Step ②: Perform wavelet transform. Use the cwt function and morlet function in the PyWavelets library of Python to calculate the wavelet coefficients of the IMF at different scales. The output of the wavelet transform is a two-dimensional matrix, where each column corresponds to a scale and each row corresponds to a time point.

[0037] Step ③: Extract features from the wavelet coefficient matrix.

[0038] The mean of wavelet coefficients: It reflects the average energy of the signal at different scales.

[0039] The variance of wavelet coefficients: It reflects the degree of fluctuation of the signal at different scales.

[0040] The peak value of wavelet coefficients: It reflects the maximum energy of the signal at different scales.

[0041] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention can achieve full-cycle fault diagnosis during the assembly and operation of agricultural tractors and has the following advantages:

[0042] (1) Full-cycle fault diagnosis can detect and repair problems at the early stage of the assembly process, effectively avoiding the accumulation of problems, thereby preventing the situation that it is difficult to detect and repair after the whole machine assembly is completed.

[0043] (2) Stage-by-stage detection ensures the gradual improvement of assembly quality, can detect and repair faults in a timely manner, reduces the rework rate after the whole machine assembly is completed, reduces maintenance costs, and improves production efficiency.

[0044] (3) By adopting the multi-sensor data fusion technology, through comprehensive analysis of the data of multiple sensors, the errors of single sensors can be eliminated, thereby improving the accuracy of fault diagnosis.

[0045] (4) The Internet of Things technology enables fault diagnosis to be carried out in real time at each stage of the assembly process, ensuring that problems can be discovered and solved in a timely manner; through the Internet of Things technology, users can remotely monitor the device status, receive fault warnings and maintenance suggestions in a timely manner, thereby improving the reliability and service life of the device.

[0046] (5) By using cloud servers and big data analysis technology, a large amount of sensor data can be efficiently processed and analyzed, thereby improving the accuracy of fault detection; based on big data analysis and machine learning algorithms, the prediction of device status and early warning of potential faults can be realized, notifying users in advance for maintenance to avoid the further expansion of faults. Brief Description of the Drawings

[0047] Figure 1 It is a schematic diagram of the detection of the fault diagnosis system of the present invention.

[0048] Figure 2 It is a schematic diagram of the sensor position setting of the present invention Figure 1 .

[0049] Figure 3 It is a schematic diagram of the sensor position setting of the present invention Figure 2 .

[0050] Figure 4 It is a flowchart of the method of the present invention.

[0051] In the figure: 1. Crankcase vibration sensor; 2. Cylinder head vibration sensor; 3. Bearing housing vibration sensor; 4. Input shaft vibration sensor; 5. Output shaft vibration sensor; 6. Gearbox vibration sensor; 7. Cylinder head temperature sensor; 8. Gearbox temperature sensor; 9. Crankcase; 10. Cylinder head; 11. Bearing housing; 12. Input shaft; 13. Output shaft; 14. Gearbox; 15. Sensor assembly; 16. Data acquisition module; 161. Engine acquisition module; 162. Gearbox acquisition module; 163. Microcontroller; 164. Data storage unit; 17. Wireless transmission module; 18. Cloud server; 19. User terminal. Specific implementation mode

[0052] The technical solution of the present invention will be further described in detail below with reference to the drawings and embodiments.

[0053] As Figures 1 - 3 shown, the present invention first provides an intelligent fault diagnosis system for an agricultural tractor based on the Internet of Things, including a sensor assembly 15, a data acquisition module 16, a wireless transmission module 17, a cloud server 18 and a user terminal 19; the above components together form: a component installation stage, a preliminary debugging stage and a comprehensive debugging stage, and the system runs through the above three stages; the sensor assembly 15 includes vibration sensors, temperature sensors and smoke sensors, wherein the vibration sensor uses an ADXL345 three-axis accelerometer, the temperature sensor selects an S18B20 digital temperature sensor, the smoke sensor selects an ATRS-1031, and the main installation locations of each sensor are on the crankcase 9, cylinder head 10, bearing housing 11, input shaft 12, output shaft 13, gearbox 14, etc., and corresponding sensors should also be installed for detection when other parts are off the production line; the data acquisition module 16 includes an engine acquisition module 161 and a gearbox acquisition module 162, and they are respectively installed near the engine, near the gearbox and on the brackets of other parts; each module includes a microcontroller 163 and a data storage unit 164, wherein the microcontroller 163 uses an ESP32 microcontroller and the data storage unit 164 uses an SD card module. The wireless transmission module 17 is a LoRa module; the cloud server 18 is linked to the whole machine detection platform, processes and analyzes the data through big data analysis and machine learning algorithms, generates fault warning information and sends it to the user terminal 19.

[0054] The vibration sensors adopt ADXL345 triaxial accelerometers and are respectively installed at the central position on the outer wall of the crankcase 9, serving as the crankcase vibration sensor 1 for monitoring the vibration signals of the crankshaft; the vibration sensor is installed at the middle position on the top of the cylinder head 10, serving as the cylinder head vibration sensor 2 for monitoring the vibration signals inside the cylinder; the vibration sensor is installed directly above the bearing housing 11, serving as the bearing housing vibration sensor 3 for monitoring the vibration signals during the operation of the bearing; the vibration sensor is installed on the outer shell near the transmission input shaft 12, serving as the input shaft vibration sensor 4 for monitoring the vibration signals of the input shaft 12; the vibration sensor is installed on the outer shell near the transmission output shaft 13, serving as the output shaft vibration sensor 5 for monitoring the vibration signals of the output shaft 13; the vibration sensor is installed on the outer shell of the gearbox 14, serving as the gearbox vibration sensor 6 for monitoring the vibration signals during the operation of the gears; the temperature sensors select S18B20 digital temperature sensors and are respectively installed on the side of the cylinder head 10 as the cylinder head temperature sensor 7 and on the outer shell of the gearbox 14 as the gearbox temperature sensor 8 for monitoring the temperature changes of the cylinder head 10 and the gearbox 14.

[0055] The data acquisition module 16 includes 163 and a data storage unit 164, and they are respectively installed on the brackets near the engine and on the brackets near the transmission; here, the data acquisition set module installed on the brackets near the engine is responsible for obtaining data from the engine sensors; the data acquisition module 16 is installed on the brackets near the transmission and is responsible for obtaining data from the transmission sensors.

[0056] The microcontroller 163 adopts an ESP32 microcontroller, which has powerful processing capabilities and wireless communication functions; the data storage unit 164 adopts an SD card module for temporarily storing sensor data to prevent data loss; the wireless transmission module 17 adopts a LoRa module to achieve long-distance data transmission and ensure that data can be transmitted to the cloud server 18 in a timely manner.

[0057] Furthermore, on the whole machine detection platform, vibration sensors and temperature sensors are installed at the same key positions of the engine, transmission and other components for monitoring the vibration signals and temperature changes during the operation of the whole machine.

[0058] Furthermore, the sensor assembly 15 cooperates through a gantry structure. When each component is produced, the gantry structure simultaneously detects each component and arranges sensors during the assembly stage of each component to ensure the assembly quality at each stage; the sensor data is collected through the acquisition module, and the data is transmitted to the cloud server 18 in real time through the wireless transmission module 17; the cloud server 18 processes and analyzes the data through big data analysis and machine learning algorithms, generates fault warning information and sends it to the user terminal 19.

[0059] This invention uses microcontrollers such as Arduino, connects vibration sensors and temperature sensors, and collects the operation data of key components of tractors in real time; these sensors can be connected to the microcontroller 163 through I2C or SPI interfaces to achieve high-precision data collection; this invention uses Google Cloud in the cloud server 18 for data storage and management, utilizes its built-in distributed database system to process big data, and calculates big data using the following method:

[0060] Step A: Construct a time-domain signal sequence X n (t) with noise, construct a white-noise time-domain signal sequence ω n (t) with a standard deviation of ε0, and combine the white-noise signal with the original signal X n (t) to construct a new signal as follows: X n (t) = X(t) + ε0ω n (t), n = 1, 2,..., K

[0061] Step B: Use EMD to decompose the constructed time-domain signal sequence X n (t) with noise, extract the first-order IMF component, and calculate the average value of its signal amplitude,

[0062]

[0063] After the first-level IMF decomposition, the residual of the signal is:

[0064]

[0065] Step C: On the basis of the residual, according to the signal processing method in step S1, reconstruct the signal r(t) + ε0EMD1(ω n (t)) by adding white noise, extract the second-order IMF component through EMD decomposition, and the calculation methods of the average value of the amplitude of this component signal and the new residual are as shown in the following formula:

[0066]

[0067] Step D: Repeat the decomposition of the signal according to the same rule, as shown in the following formula, and calculate until the residual r k (t) and the average value of the (k + 1)-order IMF until the residual r k (t) satisfies the condition of formula or the signal cannot be further decomposed by EMD,

[0068]

[0069]

[0070] where T is the length of the original signal X(t); r k (t) is the residual signal of the k-th order IMF.

[0071] As can be seen from the above formula, the signal decomposition route of CEEMDAN is the same as that of EMD. The mean curve is obtained by cubic spline interpolation fitting of the original signal value points, and then the mean curve is continuously used to iteratively screen from the upper-level signal. The mean curve used for each screening is the residual signal of the previous iteration. Therefore, it is called adaptive decomposition.

[0072] In the component installation stage, vibration sensors and temperature sensors are installed at key parts of the engine and the transmission. After the engine is installed and off the production line, pre-start is carried out to monitor the vibration and temperature changes generated during the component installation process to ensure that the components are correctly installed and meet the quality standards. The data acquisition system is used to record the sensor data in real time, and the data is transmitted to the central control system through the wireless transmission module 17. Analyze and judge whether it is normal. If yes, enter the next stage; if no, return to the component installation stage.

[0073] In the preliminary debugging stage, the positions of the sensors in the component installation stage are kept unchanged, and the installed sensors are continued to be used. The installed sensors are used to monitor the operating states of the engine and the transmission during the preliminary debugging process to initially identify potential assembly defects or faults. The vibration and temperature data during the debugging process are collected in real time, and the data is transmitted to the cloud server 18 through the wireless transmission module 17 for analysis. Analyze and judge whether it is normal. If yes, enter the next stage; if no, return to the preliminary debugging stage.

[0074] In the comprehensive debugging stage, the operating states of the whole machine are comprehensively detected to ensure that the coordinated work of all components is normal, identify and repair potential faults. The vibration and temperature data during the operation of the whole machine are collected in real time, and the data is transmitted to the cloud server 18 through the wireless transmission module 17 for analysis. Analyze and judge whether it is normal. If yes, the system debugging is completed; if no, return to the comprehensive debugging stage. After the above phased detections are completed, the overall vehicle off-line detection is carried out.

[0075] As Figure 4 shown, the present invention also provides an Internet of Things-based intelligent fault diagnosis method for agricultural tractors using the intelligent fault diagnosis system as described above, including the following steps:

[0076] Step S1: Start the simulation operation platform. During the assembly process, drive the input shafts of the engine and the gearbox through an electric motor to simulate their actual operating states. The simulation operation platform can provide a stable power source to enable each component to operate in a simulation state. The purpose of the simulation operation is to ensure the performance and state of each component during independent operation and provide basic data for subsequent detection;

[0077] Step S2: Data acquisition. Sensors collect vibration and temperature signals generated during the simulation operation in real time. The vibration sensor monitors the vibration conditions of each key part, including frequency, amplitude, and transient events. The temperature sensor monitors the temperature changes of each key part, including the temperature rise rate and the maximum temperature. The collected data is preliminarily processed by the acquisition module through the CEEMDAN algorithm and transmitted to the cloud server through the LoRa wireless transmission module.

[0078] Step S3: Data preprocessing. After receiving the data, the cloud server first performs data preprocessing. Use the Pandas library to clean the data, remove outliers and missing values, use MinMaxScaler to normalize the data to the range of 0 - 1 to ensure that different features have the same magnitude, use the Butterworth low-pass filter to filter the noise of the data, remove high-frequency noise, and retain the main components of the signal. Perform CEEMDAN decomposition on the preprocessed vibration and temperature signals to obtain several intrinsic mode functions (IMFs). Each IMF represents the oscillating components of the signal at different time scales;

[0079] Step S4: Feature extraction. Perform wavelet transform on the IMFs obtained by CEEMDAN to extract the time-frequency features of the signal. Wavelet transform is a time-frequency analysis method that can effectively detect transient events and local features in the signal. By performing wavelet transform on the IMFs obtained by CEEMDAN decomposition, the local features of the signal at different scales and positions can be obtained. Commonly used wavelet functions include Morlet wavelet, Mexican Hat wavelet, and Ricker wavelet;

[0080] Step S5: Pattern recognition and fault diagnosis. Pattern recognition and fault diagnosis using Heterogeneous Oblique Random Forest can be carried out according to the following steps:

[0081] Data preparation: Collect and preprocess the data as described in the previous step, including feature selection and data standardization,

[0082] Model training: Use the training data to construct a Heterogeneous Oblique Random Forest model. Each decision tree node uses an oblique hyperplane to divide the data to improve the classification ability of the model,

[0083] Model evaluation: Evaluate the model performance through methods such as cross-validation, and adjust parameters to optimize accuracy.

[0084] Pattern recognition and fault diagnosis: Apply the model to predict the category or state of new data, identify specific patterns or detect potential faults.

[0085] Result analysis: Analyze the model output, determine key features and patterns, and diagnose the sources of potential faults.

[0086] Step S6: Fault judgment and warning. According to the preset fault detection criteria, judge whether the vibration signal and temperature signal are abnormal. The detection criteria for the vibration signal include the enhancement of frequency components, the peak value of the vibration amplitude, etc., and the detection criteria for the temperature signal include the rising speed of the temperature and the highest temperature. When an abnormal signal is detected, the cloud server generates a fault warning message and sends a warning notification through the user terminal. The user terminal can monitor the device status in real time, receive the fault warning message in a timely manner, and provide maintenance suggestions. Specifically, the present invention performs detections at each key stage of the assembly process to ensure the gradual improvement of the assembly quality and timely discover and repair problems that occur during the assembly process.

[0087] In step S4, the Morlet wavelet is used as the mother wavelet function for feature extraction, and the specific steps are as follows:

[0088] Step ①: Select the wavelet function and scale. Select the Morlet wavelet as the mother wavelet function and determine the scale range of the wavelet transform. The present invention selects 1 to 31 as the scale range to capture the features of different frequency components.

[0089] Step ②: Perform wavelet transform. Use the cwt function and morlet function in the PyWavelets library of Python to calculate the wavelet coefficients of the IMF at different scales. The output of the wavelet transform is a two-dimensional matrix, where each column corresponds to a scale and each row corresponds to a time point.

[0090] Step ③: Extract features from the wavelet coefficient matrix.

[0091] Mean of wavelet coefficients: Reflects the average energy of the signal at different scales.

[0092] Variance of wavelet coefficients: Reflects the degree of fluctuation of the signal at different scales.

[0093] Peak value of wavelet coefficients: Reflects the maximum energy of the signal at different scales.

[0094] The machine learning model adopted by the present invention: The random forest algorithm is used for pattern recognition and fault diagnosis. Training the model: Use the Scikit-learn library to train the random forest model with historical data.

[0095] from sklearn.ensemble import RandomForestClassifier

[0096] model = RandomForestClassifier(n_estimators = 100)

[0097] model.fit(X_train, y_train)

[0098] Perform fault prediction on new data.

[0099] predictions = model.predict(X_test)

[0100] By analyzing the historical data under normal operating conditions, determine the normal range of vibration signals. If the real-time data exceeds this range, it is judged as abnormal. When the temperature exceeds the preset threshold or the temperature change rate exceeds the normal range, it is determined as a fault.

[0101] Furthermore, the present invention reduces the interference caused by other moving parts during the detection process by the following methods: Design and implement a band-pass filter through the filtering tool in MATLAB to remove low-frequency and high-frequency interference signals and only retain the vibration signals within the target frequency band. Adopt the sensor array technology, arrange multiple sensors at different positions, and use spatial filtering technology to distinguish the vibration signals of different moving parts. Install multiple vibration sensors and use the Kalman filter for data fusion to eliminate random noise and the interference of non-target moving parts.

[0102] After undergoing CEEMDAN (Complete Ensemble Empirical Mode Decomposition with Adaptive Noise), the present invention uses the Morlet wavelet as the mother wavelet function for feature extraction.

[0103] The formula for the Continuous Wavelet Transform (CWT) is as follows:

[0104]

[0105] where x(t) is the signal to be analyzed, ψ(t) is the mother wavelet function, a is the scale parameter, b is the translation parameter, and ψ * represents the complex conjugate of ψ.

[0106] Subsequently, select the wavelet function and scale. Select the Morlet wavelet as the mother wavelet function and determine the scale range of the wavelet transform. Select 1 to 31 as the scale range to capture the characteristics of different frequency components.

[0107] Perform wavelet transform:

[0108] Use the cwt function and morlet function in the PyWavelets library of Python to calculate the wavelet coefficients of the IMF at different scales. The output of the wavelet transform is a two-dimensional matrix, where each column corresponds to a scale and each row corresponds to a time point.

[0109] After that, extract features from the wavelet coefficient matrix. Select the following features for extraction:

[0110] Mean of wavelet coefficients: Reflects the average energy of the signal at different scales.

[0111] Variance of wavelet coefficients: Reflects the degree of fluctuation of the signal at different scales.

[0112] Peak value of wavelet coefficients: Reflects the maximum energy of the signal at different scales.

[0113] The present invention uses Heterogeneous Oblique Random Forest for pattern recognition and fault diagnosis. Heterogeneous Oblique Random Forest is an improved random forest algorithm that uses oblique hyperplanes instead of traditional axis-parallel decision boundaries at the nodes. This method allows considering combinations of multiple features during data partitioning, thus better capturing the geometric structure of the data and reducing computational resource consumption.

[0114] The present invention conducts fault detection and repair at each key stage of the assembly process through phased detection to ensure the assembly quality of each component; the purpose of phased detection is to serve the fault diagnosis of the final vehicle off the production line, ensuring more efficient and accurate detection after the vehicle assembly is completed; before the vehicle comes off the production line, all components have undergone phased detection and the discovered problems have been repaired; through the combination of phased detection and the final vehicle off the production line fault diagnosis, full-cycle fault diagnosis is achieved, improving the accuracy and efficiency of fault detection; phased detection discovers and repairs problems in advance, ensuring the smooth progress of the final vehicle off the production line fault diagnosis, reducing the rework rate and maintenance costs, and improving the reliability and service life of the vehicle.

[0115] The preferred embodiments of the present invention have been described in detail above, but the present invention is not limited to the specific details in the above embodiments. Within the scope of the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all fall within the protection scope of the present invention.

Claims

1. An intelligent fault diagnosis system for an agricultural tractor based on the Internet of Things, comprising a sensor assembly (15), a data acquisition module (16), a wireless transmission module (17), a cloud server (18) and a user terminal (19); characterized in that: After the above components are combined, they jointly form: a component installation stage, a preliminary debugging stage, and a comprehensive debugging stage, and the system runs through these three stages; the sensor component (15) includes a vibration sensor, a temperature sensor, and a smoke sensor. Among them, the vibration sensor uses an ADXL345 three-axis accelerometer, the temperature sensor selects an S18B20 digital temperature sensor, and the smoke sensor selects an ATRS-1031. And the main installation positions of each sensor are on the crankcase (9), cylinder head (10), bearing block (11), input shaft (12), output shaft (13), gearbox (14), etc., and corresponding sensors should also be installed for detection when other parts are off the production line; the data acquisition module (16) includes an engine acquisition module (161) and a gearbox acquisition module (162), and they are respectively installed near the engine, near the gearbox, and on the brackets of other parts; the wireless transmission module (17) is a LoRa module; the cloud server (18) is linked to the whole machine detection platform, processes and analyzes the data through big data analysis and machine learning algorithms, generates fault warning information and sends it to the user terminal (19).

2. The agricultural tractor intelligent fault diagnosis system based on the Internet of Things according to claim 1, characterized in that: The vibration sensor uses an ADXL345 three-axis accelerometer and is respectively installed at the central position on the outer wall of the crankcase (9), the middle position on the top of the cylinder head (10), directly above the bearing block (11), on the outer shell near the input shaft (12), on the outer shell near the output shaft (13), and on the outer shell of the gearbox (14); the temperature sensor selects an S18B20 digital temperature sensor and is respectively installed on the side of the cylinder head (10) and on the outer shell of the gearbox (14).

3. The intelligent fault diagnosis system for agricultural tractors based on the Internet of Things according to claim 1, characterized in that: The data acquisition module (16) includes a microcontroller (163) and a data storage unit (164), and they are respectively installed on the brackets near the engine and on the brackets near the gearbox.

4. The intelligent fault diagnosis system for agricultural tractors based on the Internet of Things according to claim 1, characterized in that: The microcontroller (163) uses an ESP32 microcontroller, which has powerful processing capabilities and wireless communication functions; the data storage unit (164) uses an SD card module to temporarily store sensor data to prevent data loss.

5. The intelligent fault diagnosis system for agricultural tractors based on the Internet of Things according to claim 1, characterized in that: The wireless transmission module (17) uses a LoRa module to achieve long-distance data transmission and ensure that data can be transmitted to the cloud server (18) in a timely manner.

6. The intelligent fault diagnosis system for agricultural tractors based on the Internet of Things according to claim 1, characterized in that: The cloud server (18) uses Google Cloud for data storage and management, uses its built-in distributed database system to process big data, and the method for processing big data is calculated as follows: Step A: Construct a time-domain signal sequence X n (t), and construct a white noise time-domain signal sequence ω n (t) with a standard deviation of ε0. Combine the white noise signal with the original signal X n (t) to construct a new signal as follows: X n (t) = X(t) + ε0ω n (t), n = 1, 2,..., K Step B: Use EMD to decompose the constructed time-domain signal sequence X with noise n (t), extract the first-order IMF component, and calculate the average value of its signal amplitude. After the first-level IMF decomposition, the residual of the signal is: Step C: Based on the residual, according to the signal processing method in Step S1, reconstruct the signal r(t)+ε0EMD1(ω n (t)) by adding white noise. Extract the second-order IMF component through EMD decomposition. The amplitude average value of this component signal and the calculation method of the new residual are shown in the following formula: Step D: Repeat the signal decomposition according to the same rules, as shown in the following formula, and calculate until the residual r of the k-th order IMF k (t) and the average value of the (k + 1)-th order IMF until the residual r k (t) satisfies the condition of formula or until the signal cannot be further decomposed by EMD where T is the length of the original signal X(t); r k (t) is the residual signal of the k-th order IMF.

7. The intelligent fault diagnosis system for agricultural tractors based on the Internet of Things according to claim 1, characterized in that: In the component installation stage, vibration sensors and temperature sensors are installed at key parts of the engine and the gearbox; in the preliminary debugging stage, the positions of the sensors in the component installation stage remain unchanged, and the installed sensors are continued to be used; in the comprehensive debugging stage, the operating status of the whole machine is comprehensively detected to ensure the normal collaborative work of all components, identify and repair potential faults.

8. An intelligent fault diagnosis method for an agricultural tractor based on the Internet of Things as described in any one of claims 1-6, characterized in that: It includes the following steps: Step S1: Start the simulation running platform. During the assembly process, drive the input shaft (12) of the engine and the transmission through the motor to simulate their actual operating states. The simulation running platform can provide a stable power source to enable each component to operate in a simulation state. The purpose of the simulation running is to ensure the performance and state of each component during independent operation and provide basic data for subsequent detection; Step S2: Data acquisition. Sensors collect vibration and temperature signals generated during the simulation running in real time. The vibration sensor monitors the vibration conditions of each key part, including frequency, amplitude, and transient events. The temperature sensor monitors the temperature changes of each key part, including the temperature rise rate and the maximum temperature. The collected data is preliminarily processed by the acquisition module through the CEEMDAN algorithm and transmitted to the cloud server through the LoRa wireless transmission module. Step S3: Data preprocessing. After the cloud server receives the data, it first performs data preprocessing. Use the Pandas library to clean the data, remove outliers and missing values. Use MinMaxScaler to normalize the data to the range of 0-1 to ensure that different features have the same magnitude. Use the Butterworth low-pass filter to filter the noise of the data, remove high-frequency noise, and retain the main components of the signal. Perform CEEMDAN decomposition on the preprocessed vibration and temperature signals to obtain several intrinsic mode functions (IMFs). Each IMF represents the oscillatory component of the signal at different time scales; Step S4: Feature extraction. Perform wavelet transform on the IMFs obtained by CEEMDAN to extract the time-frequency features of the signal. Wavelet transform is a time-frequency analysis method that can effectively detect transient events and local features in the signal. By performing wavelet transform on the IMFs obtained by CEEMDAN decomposition, the local features of the signal at different scales and positions can be obtained. Commonly used wavelet functions include Morlet wavelet, Mexican Hat wavelet, and Ricker wavelet; Step S5: Pattern recognition and fault diagnosis. Pattern recognition and fault diagnosis using Heterogeneous Oblique Random Forest can be carried out according to the following steps: Data preparation: Collect and preprocess the data as described in the previous step, including feature selection and data standardization, Model training: Use the training data to construct a Heterogeneous Oblique Random Forest model. Each decision tree node uses an oblique hyperplane to divide the data to improve the classification ability of the model, Model evaluation: Evaluate the model performance through methods such as cross-validation and adjust the parameters to optimize the accuracy, Pattern recognition and fault diagnosis: Apply the model to predict the category or state of new data, identify specific patterns or detect potential faults, Result analysis: Analyze the model output, determine the key features and patterns, and diagnose the potential sources of faults; Step S6: Fault judgment and warning. According to the preset fault detection criteria, judge whether the vibration signal and temperature signal are abnormal. The detection criteria for the vibration signal include the enhancement of frequency components, the peak value of the vibration amplitude, etc. The detection criteria for the temperature signal include the rising speed of the temperature and the highest temperature. When an abnormal signal is detected, the cloud server generates a fault warning message and sends a warning notification through the user terminal. The user terminal can monitor the device status in real time, receive the fault warning message in a timely manner, and provide maintenance suggestions. Specifically, in the present invention, detection is carried out at each key stage of the assembly process to ensure the gradual improvement of the assembly quality and timely discovery and repair of problems occurring in the assembly process.

9. The intelligent fault diagnosis method for agricultural tractors based on the Internet of Things according to claim 8, characterized in that: In step S4, the Morlet wavelet is used as the mother wavelet function for feature extraction. The specific steps are as follows: Step ①: Select the wavelet function and scale. Select the Morlet wavelet as the mother wavelet function and determine the scale range of the wavelet transform. In the present invention, 1 to 31 is selected as the scale range to capture the characteristics of different frequency components; Step ②: Perform wavelet transform. Use the cwt function and morlet function in the PyWavelets library of Python to calculate the wavelet coefficients of the IMF at different scales. The output of the wavelet transform is a two-dimensional matrix, where each column corresponds to a scale and each row corresponds to a time point; Step ③: Extract features from the wavelet coefficient matrix. The mean value of the wavelet coefficients: Reflects the average energy of the signal at different scales. The variance of the wavelet coefficients: Reflects the degree of fluctuation of the signal at different scales. The peak value of the wavelet coefficients: Reflects the maximum energy of the signal at different scales.

Images