Maintenance inspection system for speed reducer based on multi-source sensing fusion

Abstract

The application relates to a multi-source sensing fusion-based maintenance inspection system for a speed reducer, which comprises a gear speed reducer, a data detection device, a data acquisition subsystem, a communication subsystem, a data analysis subsystem, an intelligent diagnosis subsystem, a three-axis vibration sensor and a microphone arranged in an acoustic array, forms a Cartesian coordinate system, can accurately capture fault information from two dimensions of vibration and acoustics, greatly improves the detection sensitivity of early faults, and effectively reduces detection delay.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence, and in particular to a reducer maintenance and inspection system based on multi-source sensor fusion. Background Art

[0002] Existing online detection systems mostly rely on sensors such as vibration, oil, and temperature, and perform preliminary fusion of multi-source sensors through weighted averaging, Kalman filtering, or neural networks. For example, vibration and temperature are jointly detected, mechanical faults are analyzed through vibration spectrum, and temperature sensors are used to assist in warning of overheating risks. This solution has high latency and is not sensitive enough for early fault detection. Some also use inductive or optical sensors to detect oil particles, judge the degree of wear, and realize partial condition prediction. There are also Hall-effect speed sensors that assist vibration sensors in detection through time domain interpolation algorithms, but there is a problem of phase error accumulation.

[0003] At the same time, the sensor maintains high-frequency data acquisition, resulting in excessive power consumption, and the huge amount of data puts great pressure on the bandwidth; the existing fault defects and vague fault sources make it difficult for traditional vibration sensors to distinguish between bearing, gear and shaft faults, especially in vibration scenarios with multiple components coupled, with a high misjudgment rate; the calculation of fault characteristic frequency depends on the speed accuracy, and the speed sensor is not integrated, resulting in spectrum analysis errors; there is a lack of acoustic and vibration joint positioning capabilities, and the physical location of the faulty component cannot be determined. Oil, temperature and vibration data are analyzed independently, and no cross-parameter correlation model has been established. Therefore, there is still room for improvement in the existing technology. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a reducer maintenance and inspection system based on multi-source sensor fusion in response to the above-mentioned deficiencies in the prior art.

[0005] To achieve the above-mentioned objectives, the present invention provides the following technical solution: a reducer maintenance and inspection system based on multi-source sensor fusion, comprising a gear reducer, a data detection device, a data acquisition subsystem, a communication subsystem, a data analysis subsystem, and an intelligent diagnosis subsystem, characterized in that: the gear reducer comprises a housing, in which an input shaft, an output shaft, a high-speed shaft and a low-speed shaft are rotatably arranged, and the low-speed shaft, the high-speed shaft, the input shaft and the output shaft are connected by a gear set.

[0006] The data detection equipment includes a three-axis vibration sensor, a microphone, a temperature sensor, a Hall speed sensor, a multi-parameter oil sensor, an adsorption-type wear particle sensor, and a water content sensor. The three-axis vibration sensor is respectively installed at the connection between the box body and the input shaft and the output shaft. The microphone is arranged on the outer periphery of the box body through an acoustic array to detect the internal noise of the gearbox, facilitate the location of faulty components and the analysis of fault problems. The installation position of the three-axis vibration sensor and the microphone forms a Cartesian coordinate system, and the Cartesian coordinate system is coplanar with the low-speed shaft, high-speed shaft, input shaft and output shaft. The temperature sensor is distributed on the high-speed shaft, low-speed shaft, input shaft and the connection between the output shaft and the box body. The Hall speed sensor is fixedly installed on the input shaft. The multi-parameter oil sensor, adsorption-type wear particle sensor, water content sensor and are arranged on the side of the box body away from the output shaft.

[0007] The data acquisition subsystem is used to collect and integrate data from data detection equipment;

[0008] The communication subsystem transmits the collected data to the data analysis system and the intelligent diagnosis system by wireless communication;

[0009] The data analysis subsystem includes receiving data from the data acquisition subsystem and calculating the fault coordinates;

[0010] The intelligent diagnosis subsystem includes a sub-database, and the intelligent diagnosis subsystem uses the fault coordinates calculated by the data analysis subsystem to perform fault judgment on the component.

[0011] The above technical solution is adopted, through multi-source sensor fusion, using three-axis vibration sensors, microphones, temperature sensors, Hall speed sensors, multi-parameter oil sensors, adsorption wear sensors, moisture content sensors and other sensors to work together. The three-axis vibration sensor and the microphones arranged in an acoustic array form a Cartesian coordinate system, which can accurately capture fault information from both vibration and acoustic dimensions, greatly improving the detection sensitivity of early faults and effectively reducing detection delays. The multi-parameter oil sensor acts as a lubrication environment detector and auxiliary judgment. The multi-parameter oil sensor includes detection of temperature, density, and viscosity. The oil is stirred inside the box. The oil sensor will detect the corresponding state of the oil. If the oil viscosity drops to a certain value and the box temperature rises, a corresponding alarm signal will be issued. If faults such as bearing or gear wear are determined in the previous steps, a wear condition assessment will be performed based on the changes in the density of metal particles in the oil and the noise and vibration changes of the gearbox over a period of time, and it is easier to diagnose. It is easy to judge the wear condition. When the density suddenly changes, it may be caused by problems such as external water ingress, and corresponding judgments can be made. The oil sensor uses multi-physical quantity collaborative detection to accurately evaluate the lubricating oil status and wear trends of various parts of the gearbox. The intelligent diagnosis subsystem of this system combines the fault coordinates calculated by the data analysis subsystem and uses the sub-database to make fault judgments, which can effectively avoid these errors and improve the accuracy of fault diagnosis. The existing system vibration sensor has a high misjudgment rate in distinguishing bearing, gear, and shaft faults, especially in multi-component coupled vibration scenarios. The system's multi-sensor collaboration and acoustic and vibration joint positioning can accurately determine the physical location of the faulty component, greatly reducing the misjudgment rate. The system uses multi-source sensor fusion to comprehensively analyze the data of different types of sensors, and establishes a more comprehensive and in-depth correlation model, which can more accurately reflect the overall operating status of the reducer, realize accurate diagnosis and predictive maintenance of reducer faults, reduce equipment failure rate and maintenance costs, and improve equipment reliability and operating efficiency.

[0012] The above-mentioned reducer maintenance and inspection system based on multi-source sensor fusion can be further configured as follows: the three-axis vibration sensor and the microphone form a Cartesian coordinate system with the corresponding relationship being: the microphones are respectively installed on the X-axis, Y-axis and the coordinate origin, and the three-axis vibration sensors are respectively installed on the X-axis and Y-axis to make the coordinate plane coplanar with the gear shaft axis, and all fault sources are within the coordinate system.

[0013] Using the above technical solution, the axis plane perpendicular to the output shaft is the x-axis, and the axis perpendicular to the input shaft is the y-axis. The x-axis and y-axis intersect at the microphone installation position, which is the coordinate origin. Microphones are installed on the x-axis, y-axis, and coordinate origin respectively. Vibration sensors are installed on the x-axis and y-axis so that the coordinate plane is coplanar with the gear shaft axis. All fault sources are within the coordinate system. The microphones distributed on the coordinate axes and the origin can capture noise information inside the gearbox from multiple directions and positions, effectively collecting noise generated at different locations and more accurately locating the fault source. When a fault occurs, the approximate area of ​​the fault can be quickly determined by analyzing the time difference of the sound signals received by different microphones. The vibration sensors installed on the x-axis and y-axis can effectively detect the vibration of the gear and bearing. Because the coordinate plane is coplanar with the gear shaft axis, it can maximize the acquisition of vibration information related to the operating status of the gear shaft, improving the sensitivity and accuracy of the vibration signal. Combining acoustic and vibration monitoring within a Cartesian coordinate system can achieve synchronous analysis of acoustic and vibration data.

[0014] The above-mentioned reducer maintenance and inspection system based on multi-source sensor fusion can be further configured as follows: the data analysis subsystem algorithm specifically includes the following steps:

[0015] Step S1: Data preparation, receiving data from the data acquisition subsystem;

[0016] Step S2: Determine the location coordinates (x, y) of the fault source by the time difference between the sound propagation times to different locations. The formula used is as follows:

[0017] : It represents the time it takes for the sound to propagate from the fault source S(x,y) to the origin (or reference point);

[0018] d1 refers to the distance between the microphone and the origin coordinate on the X axis;

[0019] : d2 is the distance between the microphone and the origin coordinate on the Y axis;

[0020] Among them, the speed of sound in air is vair=343m / s;

[0021] Step S3: Estimate the time delay difference by using the generalized cross-correlation phase transform algorithm, and use the time difference to establish and solve the hyperbola equation to obtain the signal source;

[0022] Step S4: Obtain the coordinates of the fault source through the Newton iteration method, and further determine the fault source location through particle filter fusion optimization, associate multiple database data with the fault source state, and then the fault source location changes over time. In the particle filter iteration process, initialization is first performed, and the generated particles obey , and then update the weights;

[0023] Step S5: Resample to retain high-weight particles and perform weighted averaging to obtain the final coordinates:

[0024]

[0025] in, Refers to the weight of the i-th particle at time k, is the actual observed value, Refers to the predicted observation value of the i-th particle.

[0026] Using the above technical solution, during the data preparation phase, data is received from the data acquisition subsystem, providing a foundation for subsequent analysis and ensuring data integrity and accuracy. The fault source coordinates are determined by the time difference between sound propagation to different locations. A formula is used to calculate the time it takes for sound to propagate to the origin, the first sensor, and the second sensor. Combined with the known speed of sound in air and solids, the fault source can be initially located. A generalized cross-correlation phase transform algorithm estimates the time difference and establishes and solves a hyperbolic equation, further narrowing the fault source range and improving positioning accuracy. The Newton iteration method is used to obtain the fault source coordinates, which are then further determined through particle filter fusion optimization. This method leverages the advantages of the Newton iteration method, such as its fast convergence speed. The particle filter also uses the particle filter to fuse data from multiple databases with the fault source state and the temporal variation of the fault source location, making positioning more accurate and dynamic. Resampling retains high-weight particles and performs weighted averaging to obtain the final coordinates, effectively removing interference from low-weight particles and making the final fault source coordinates more reliable and stable.

[0027] The above-mentioned speed reducer maintenance and inspection system based on multi-source sensor fusion can be further configured as follows: Step S3 includes Step S3.1: estimating the delay difference by using a generalized cross-correlation phase transformation algorithm, and correcting the delay difference. The formula used is as follows:

[0028] ;

[0029] Where Δtcomp is the corrected delay difference, dwall is the box wall thickness, and vwall = 5000 m / s is the speed of sound in the solid (box wall). dair is the distance of the air part in the box, vair is the speed of sound in air;

[0030] Step S3.2 uses the time difference to establish and solve the hyperbola equation to obtain the signal source. The hyperbola equation is:

[0031] ;

[0032] ;

[0033] Where t10=t1−t0, t1 is the time required for the signal to propagate from the target point to the reference point (d1,0), t0 is the time required for the signal to propagate from the target point to the origin (0,0), and Δt10 is the difference between these two propagation times;

[0034] Δt20=t2−t0, where t2 is the time required for the signal to propagate from the target point to the reference point (0, d2), t0 is the time required for the signal to propagate from the target point to the origin (0, 0), and Δt20 is the difference between these two propagation times.

[0035] Using the above technical solution, in step S3.1, the delay difference is estimated by the generalized cross-correlation phase transformation algorithm, and the formula is used The time delay difference is corrected. Considering that sound propagates through two different media—the enclosure wall and the air within it—and that the speed of sound varies in these media, the correction formula fully accounts for factors such as the enclosure wall thickness, the speed of sound in the solid (enclosure wall), the distance between the air portions within the enclosure, and the speed of sound in the air. This makes the calculation of the time delay difference more accurate and eliminates errors caused by media differences. In step S3.2, the time difference is used to establish and solve a hyperbolic equation. Based on the time difference in sound propagation, the location of the signal source can be linked to the location of the sensor, as well as the speed of sound propagation and the time difference. By solving these two hyperbolic equations, the signal source can be located, accurately locating the fault source.

[0036] The above-mentioned reducer maintenance and inspection system based on multi-source sensor fusion can be further configured as follows: Step S4 includes step S4.1: obtaining the coordinates of the fault source through the Newton iteration method, and further determining the location of the fault source through particle filter fusion optimization, and associating multiple databases with the data fault source status. The formula used is as follows:

[0037]

[0038] in is the state vector, containing Fault source coordinates at time and its rate of change , is the state transition matrix, is the process noise;

[0039] Step S4.2: The subsequent fault source location changes over time and the observation equation is listed:

[0040]

[0041] in, is the observation vector, including RMS, which is the root mean square value of the vibration signal. is the bearing temperature rise rate (unit: °C / min), is the acoustic initial value coordinate, H is the observation matrix, and the state vector is mapped to the observation space. is the observation noise, which obeys the Gaussian distribution N(0,R), R is the noise covariance matrix;

[0042] Step S4.3: During the particle filter iteration process, initialization is first performed and the generated particles obey , and then update the weights. The weight update formula is:

[0043] ;

[0044] Using the above technical solution, step S4.1 uses the Newton iteration method to obtain the coordinates of the fault source. This method can quickly converge to the approximate value of the fault source coordinates, laying the foundation for subsequent accurate analysis. The fault source location is further determined through particle filter fusion optimization, and multiple database data are associated with the fault source state, making full use of the state vector Describe the coordinates of the fault source and its rate of change, combined with the state transition matrix F and process noise , so that the determination of the fault source location is more consistent with the actual physical process, taking into account the dynamic change characteristics of the fault source, and enhancing the accuracy and reliability of positioning. Step S4.2 lists the observation equation and transforms the observation vector The vibration signal root mean square value, bearing temperature rise rate and acoustic initial value coordinates in the state vector are associated with the state vector through the observation matrix H, while considering the observation noise. , so that the system can analyze the fault source location by integrating multi-source data, and improve the ability to perceive the fault source state; Step S4.3 In the particle filter iteration process, the initialization generates The particle weights are updated through the weight update formula, which can dynamically adjust the particle weights according to the difference between the actual observation value and the predicted observation value, retain the particles that are more consistent with the actual situation, remove the interference information, and make the final fault source location more accurate.

[0045] The above-mentioned reducer maintenance and inspection system based on multi-source sensor fusion can be further configured as follows: the algorithm of the intelligent diagnosis subsystem specifically includes the following steps:

[0046] Step Q1: After obtaining the coordinates according to the data analysis subsystem, the component is judged by first roughly comparing the samples and then loading the sub-database according to the coordinates.

[0047] Step Q2: Use the dynamic time warping method to align the real-time features and samples and calculate the comprehensive similarity. The calculation formula is:

[0048] ;

[0049] in, is the real-time vibration eigenvector, is the database sample feature vector, Temperature feature similarity (normalized difference), Acoustic positioning coordinate matching: If S ≥ 0.95, the fault type is directly determined. If 0.7 ≤ S ≤ 0.9, the data is uploaded to the cloud, triggering secondary verification.

[0050] Using the above technical solution, in step Q1, a rough sample comparison is performed based on the coordinates obtained by the data analysis subsystem. The sub-database is then loaded based on the coordinates, rapidly narrowing the search scope and accurately locating the relevant data of the potentially faulty component. In step Q2, dynamic time warping (DTW) is used to align real-time features with samples and calculate a comprehensive similarity. This formula comprehensively considers the real-time vibration feature vector, database sample feature vector, temperature feature similarity, and acoustic location coordinate matching, evaluating fault characteristics from multiple dimensions. This avoids the limitations of single-feature judgment and improves the accuracy of fault diagnosis. Different diagnostic strategies are developed based on the value of the comprehensive similarity S. When S ≥ 0.95, the fault type is directly determined, enabling rapid processing of faults with high similarity and distinct features, improving diagnostic efficiency and enabling timely implementation of appropriate maintenance measures. When S ≤ 0.7 ≤ 0.9, data is uploaded to the cloud to trigger secondary verification. For cases with moderate similarity and unclear fault characteristics, further analysis is performed using the cloud's powerful computing power and abundant data resources to ensure the reliability of the diagnostic results.

[0051] The above-mentioned reducer maintenance and inspection system based on multi-source sensor fusion can be further configured as follows: Step Q1 includes step Q1.1: using PCA dimensionality reduction technology to reduce the 32-dimensional vibration feature to 8 dimensions. The dimensionality reduction formula is:

[0052] ;

[0053] Among them, V is the original 32-dimensional feature vector, P is the principal component matrix;

[0054] Step Q1.2: Use INT8 quantization technology to convert 32-bit floating-point numbers (FP32) into 8-bit integers (INT8) to reduce the amount of computation and speed up inference. The quantization formula is:

[0055]

[0056] Here, μ is the data mean, σ is the data standard deviation, and FP32 floating-point numbers are compressed to INT8 integers, reducing the model memory usage from 32MB to 8MB and shortening the inference time.

[0057] Using the above technical solution, step Q1.1 uses PCA dimensionality reduction technology to reduce the 32-dimensional vibration features to 8 dimensions, effectively removing redundant information from the original features. This dimensionality reduction formula not only reduces the data dimension but also maximizes the preservation of key data information, reducing the complexity of subsequent calculations and improving computational efficiency. Step Q1.2 uses INT8 quantization technology to convert 32-bit floating-point numbers (FP32) to 8-bit integers (INT8), significantly reducing the amount of computation and accelerating inference. This is because 8-bit integer operations are more efficient on hardware than 32-bit floating-point operations. Furthermore, the model memory usage is reduced from 32MB to 8MB, significantly saving memory resources and enabling the system to run more smoothly in resource-constrained environments.

[0058] The above-mentioned reducer maintenance and inspection system based on multi-source sensor fusion can be further configured as follows: Step Q2 includes Step Q2.1: In the interval of secondary verification (0.7≤S≤0.9), the system will immediately upload the relevant data to the cloud, and further determine the type, severity and possible development trend of the fault through the cloud.

[0059] Step Q2.1: The cloud feeds back the judgment results to the local device and provides corresponding maintenance suggestions based on the fault situation.

[0060] Using the above technical solution, the system uploads relevant data to the cloud during the secondary verification interval (0.7≤S≤0.9). This measure fully utilizes the cloud's powerful computing power and rich data resources. The cloud can conduct a more in-depth and comprehensive analysis of the fault based on a large amount of historical data and advanced algorithm models, thereby further determining the type, severity, and possible development trend of the fault. This makes up for the shortcomings of local equipment in data processing and analysis capabilities, and improves the accuracy and reliability of fault diagnosis. The cloud then feeds back the judgment results to the local device and provides corresponding maintenance recommendations based on the fault situation, allowing local equipment operators to handle the fault in a timely manner. Throughout the monitoring and fault handling process, the system will continue to self-learn and optimize. By accumulating and analyzing a large amount of equipment operation data, the system can automatically adjust the sampling frequency and operating status of the sensor. For devices operating stably and in good condition, some sensors can reduce their sampling frequency or enter a dormant state at appropriate times, reducing energy consumption and extending sensor lifespan. For devices experiencing abnormal conditions, the system automatically increases the sensor sampling frequency to more closely monitor device status and promptly capture details of developing faults. Monitoring data from different devices is aggregated and analyzed in the cloud. Through technologies such as federated learning, devices can share fault knowledge and handling experience. When a device detects a new fault mode, the relevant information is collated and analyzed in the cloud, and a more robust judgment algorithm is shared over-the-air (OTA) with other devices. This continuously improves the fault prediction and diagnosis capabilities of the entire system, enabling more efficient and intelligent predictive maintenance.

[0061] The above-mentioned speed reducer maintenance and inspection system based on multi-source sensor fusion can be further configured as follows: the input shaft is arranged perpendicular to the high-speed shaft, and the high-speed shaft, low-speed shaft and output shaft are arranged in parallel.

[0062] By adopting the above technical solution, the structure in which the input shaft is perpendicular to the high-speed shaft can realize the change of power direction through appropriate transmission components, so that the power can be smoothly transmitted from the input shaft to the high-speed shaft, and the high-speed shaft, low-speed shaft and output shaft are arranged in parallel, which is conducive to the use of a common and efficient parallel shaft gear transmission method. This transmission method has high transmission efficiency and strong load-bearing capacity, and can effectively transmit the power of the high-speed shaft to the low-speed shaft and output shaft step by step, reducing energy loss during power transmission.

[0063] The above-mentioned reducer maintenance and inspection system based on multi-source sensor fusion can be further configured as follows: the gear group includes a first rotating gear arranged on the output shaft toward one end of the high-speed shaft, the high-speed shaft is provided with a second transmission gear meshing with the first transmission gear, the high-speed shaft is provided with a third transmission gear located on one side of the second transmission gear, the low-speed shaft is provided with a fourth transmission gear meshing with the third transmission gear, the low-speed shaft is provided with a fifth transmission gear located on one side of the fourth transmission gear, and the output shaft is provided with a sixth transmission gear meshing with the fifth transmission gear.

[0064] By adopting the above technical solution, the power of the output shaft can be smoothly and efficiently transmitted to the high-speed shaft through the engagement of the first rotating gear on the output shaft with the second transmission gear on the high-speed shaft, thereby realizing the initial conversion and transmission of power. The third transmission gear on the high-speed shaft is engaged with the fourth transmission gear on the low-speed shaft, and the fifth transmission gear on the low-speed shaft is engaged with the sixth transmission gear on the output shaft, forming a multi-stage transmission structure to meet the requirements of speed and torque under different working conditions, so that the reducer can adapt to a wider range of working scenarios.

[0065] Beneficial effects of the present invention:

[0066] 1. A Cartesian coordinate system is formed by a three-axis vibration sensor and microphones arranged in an acoustic array, which can accurately capture fault information from both vibration and acoustic dimensions, greatly improving the sensitivity of early fault detection and effectively reducing detection delays.

[0067] 2. The multi-parameter oil state detection sensor analyzes changes in oil state caused by internal and external conditions, improving the recognition rate of complex faults and thus establishing a vibration-acoustics-temperature-oil joint degradation model.

[0068] 3. It can dynamically adjust the sensor frequency according to the algorithm to reduce the sensor power consumption and the amount of data generated. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 Schematic diagram of the gear reducer structure of the present invention;

[0070] Figure 2 It is a side view of the gear reducer structure of the present invention;

[0071] Figure 3 This is a structural diagram of the gear set of the present invention;

[0072] Figure 4 Schematic diagram of the system flow of the present invention;

[0073] Figure 5 Creating a Cartesian coordinate diagram for the present invention;

[0074] Label notes: 1-gear reducer, 2-housing, 3-input shaft, 4-output shaft, 5-high-speed shaft, 6-low-speed shaft, 7-three-axis vibration sensor, 8-microphone, 9-temperature sensor, 10-Hall speed sensor, 11-multi-parameter oil sensor, 12-adsorption abrasive sensor, 13-water content sensor, 14-first rotating gear, 15-second transmission gear, 16-third transmission gear, 17-fourth transmission gear, 18-fifth transmission gear, 19-sixth transmission gear. DETAILED DESCRIPTION

[0075] This specific embodiment is merely an explanation of the present invention and is not intended to limit the present invention. After reading this specification, those skilled in the art may make non-creative modifications to this embodiment as needed. However, as long as such modifications are within the scope of the claims of the present invention, they are protected by patent law. Example 1

[0076] The reducer maintenance and inspection system based on multi-source sensor fusion includes a gear reducer 1, data detection equipment, a data acquisition subsystem, a communication subsystem, a data analysis subsystem, and an intelligent diagnosis subsystem, and is characterized in that the gear reducer includes a housing 2, in which an input shaft 3, an output shaft 4, a high-speed shaft 5 and a low-speed shaft 6 are rotatably arranged, and the low-speed shaft 6, the high-speed shaft 5, the input shaft 3 and the output shaft 4 are connected by a gear set.

[0077] The data detection equipment includes a three-axis vibration sensor 7, a microphone 8, a temperature sensor 9, a Hall speed sensor 10, a multi-parameter oil sensor 11, an adsorption-type wear particle sensor 12, and a water content sensor 13. The three-axis vibration sensor 7 is respectively installed at the connection between the box body 2 and the input shaft 3 and the output shaft 4. The microphone 8 is arranged on the periphery of the box body 2 through an acoustic array to detect the internal noise of the gearbox, facilitate the location of faulty components and the analysis of fault problems. The installation positions of the three-axis vibration sensor 7 and the microphone 8 form a Cartesian coordinate system, which is coplanar with the low-speed shaft 6, the high-speed shaft 5, the input shaft 3 and the output shaft 4. The temperature sensor 9 is distributed at the connection between the high-speed shaft 5, the low-speed shaft 6, the input shaft 3 and the output shaft 4 and the box body 2. The Hall speed sensor 10 is fixedly installed on the input shaft 3. The multi-parameter oil sensor 11, the adsorption-type wear particle sensor 12, the water content sensor 13 and the box body 2 are arranged on the side away from the output shaft.

[0078] The data acquisition subsystem is used to collect and integrate data from data detection equipment;

[0079] The communication subsystem uses wireless communication to transmit the collected data to the data analysis system and intelligent diagnosis system;

[0080] The data analysis subsystem includes receiving data from the data acquisition subsystem and calculating the fault coordinates;

[0081] The intelligent diagnosis subsystem includes a sub-database, and the intelligent diagnosis subsystem uses the fault coordinates calculated by the data analysis subsystem to perform fault judgment on the component.

[0082] The three-axis vibration sensor 7 and the microphone 8 form a Cartesian coordinate system with the corresponding relationship as follows: the microphone 8 is respectively installed on the X-axis, Y-axis and the coordinate origin, and the three-axis vibration sensor 7 is respectively installed on the X-axis and Y-axis to make the coordinate plane coplanar with the gear shaft axis, and all fault sources are inside the coordinate system. Example 2

[0083] The data analysis subsystem algorithm specifically includes the following steps:

[0084] Step S1: Data preparation, receiving data from the data acquisition subsystem;

[0085] Step S2: Determine the location coordinates (x, y) of the fault source by the time difference between the sound propagation times to different locations. The formula used is as follows:

[0086] : It represents the time it takes for the sound to propagate from the fault source S(x,y) to the origin (or reference point);

[0087] d1 refers to the distance between the microphone and the origin coordinate on the X axis;

[0088] : d2 is the distance between the microphone and the origin coordinate on the Y axis;

[0089] Among them, the speed of sound in air is vair=343m / s;

[0090] Step S3: Estimate the time delay difference by using the generalized cross-correlation phase transform algorithm, and use the time difference to establish and solve the hyperbola equation to obtain the signal source;

[0091] Step S4: Obtain the coordinates of the fault source through the Newton iteration method, and further determine the fault source location through particle filter fusion optimization, associate the fault source status of multiple official data, and the subsequent fault source location changes over time. In the particle filter iteration process, initialization is first performed, and the generated particles obey , and then update the weights;

[0092] Step S5: Resample to retain high-weight particles and perform weighted averaging to obtain the final coordinates:

[0093]

[0094] in, Refers to the weight of the i-th particle at time k, is the actual observed value, Refers to the predicted observation value of the i-th particle.

[0095] Step S3 includes step S3.1: estimating the delay difference by using a generalized cross-correlation phase transform algorithm, and correcting the delay difference. The formula used is as follows:

[0096] ;

[0097] Among them, Δtcomp is the corrected delay difference, dwal l is the box wall thickness, vwall=5000m / s is the speed of sound in the solid (box wall), dair is the distance of the air part in the box, vair is the speed of sound in air;

[0098] Step S3.2 uses the time difference to establish and solve the hyperbola equation to obtain the signal source. The hyperbola equation is:

[0099] ;

[0100] ;

[0101] Where t10=t1−t0, t1 is the time required for the signal to propagate from the target point to the reference point (d1,0), t0 is the time required for the signal to propagate from the target point to the origin (0,0), and Δt10 is the difference between these two propagation times;

[0102] Δt20=t2−t0, where t2 is the time required for the signal to propagate from the target point to the reference point (0, d2), t0 is the time required for the signal to propagate from the target point to the origin (0, 0), and Δt20 is the difference between these two propagation times.

[0103] Step S4 includes step S4.1: obtaining the coordinates of the fault source by the Newton iteration method, further determining the location of the fault source by particle filter fusion optimization, and associating multiple database data with the fault source status. The formula used is as follows:

[0104]

[0105] in is the state vector, containing Fault source coordinates at time and its rate of change , is the state transition matrix, is the process noise;

[0106] Step S4.2: The subsequent fault source location changes over time and the observation equation is listed:

[0107]

[0108] in, is the observation vector, including RMS, which is the root mean square value of the vibration signal. is the bearing temperature rise rate (unit: °C / min), is the acoustic initial value coordinate, H is the observation matrix, and the state vector is mapped to the observation space. is the observation noise, which obeys the Gaussian distribution N(0,R), R is the noise covariance matrix;

[0109] Step S4.3: During the particle filter iteration process, initialization is first performed and the generated particles obey , and then update the weights. The weight update formula is:

[0110] ; Example 3

[0111] The intelligent diagnosis subsystem algorithm specifically includes the following steps:

[0112] Step Q1: After obtaining the coordinates according to the data analysis subsystem, the component is judged by first roughly comparing the samples and then loading the sub-database according to the coordinates.

[0113] Step Q2: Use the dynamic time warping method to align the real-time features and samples and calculate the comprehensive similarity. The calculation formula is:

[0114] ;

[0115] in, is the real-time vibration eigenvector, is the database sample feature vector, Temperature feature similarity (normalized difference), Acoustic positioning coordinate matching: If S ≥ 0.95, the fault type is directly determined. If 0.7 ≤ S ≤ 0.9, the data is uploaded to the cloud, triggering secondary verification.

[0116] Step Q1 includes step Q1.1: Using PCA dimensionality reduction technology, the 32-dimensional vibration feature is reduced to 8 dimensions. The dimensionality reduction formula is:

[0117] ;

[0118] Among them, V is the original 32-dimensional feature vector, P is the principal component matrix;

[0119] Step Q1.2: Use INT8 quantization technology to convert 32-bit floating-point numbers (FP32) into 8-bit integers (INT8) to reduce the amount of computation and speed up inference. The quantization formula is:

[0120]

[0121] Here, μ is the data mean, σ is the data standard deviation, and FP32 floating-point numbers are compressed to INT8 integers, reducing the model memory usage from 32MB to 8MB and shortening the inference time.

[0122] Step Q2 includes step Q2.1: Secondary verification interval (0.7≤S≤0.9). The system will immediately upload the relevant data to the cloud, through which the type, severity, and possible development trend of the fault can be further determined.

[0123] Step Q2.1: The cloud feeds back the judgment results to the local device and provides corresponding maintenance suggestions based on the fault situation. Example 4

[0124] The input shaft 3 of the gear reducer 1 is arranged perpendicularly to the high-speed shaft 5 , and the high-speed shaft 5 , the low-speed shaft 6 and the output shaft 4 are arranged in parallel.

[0125] The gear set includes a first rotating gear 14 arranged at one end of the output shaft 4 facing the high-speed shaft 5, the high-speed shaft 5 is provided with a second transmission gear 15 meshing with the first transmission gear, the high-speed shaft 5 is provided with a third transmission gear 16 on one side of the second transmission gear 15, the low-speed shaft 6 is provided with a fourth transmission gear 17 meshing with the third transmission gear 16, and the low-speed shaft 6 is provided with a fifth transmission gear 18 on one side of the fourth transmission gear 17. The output shaft 4 is provided with a sixth transmission gear 19 meshing with the fifth transmission gear 18. Through the meshing of the first rotating gear 14 on the output shaft 4 and the second transmission gear 15 on the high-speed shaft 5, the power of the output shaft 4 can be smoothly and efficiently transmitted to the high-speed shaft 5, realizing the initial conversion and transmission of power. The third transmission gear 16 on the high-speed shaft 5 is meshed with the fourth transmission gear 17 on the low-speed shaft 6, and the fifth transmission gear 18 on the low-speed shaft 6 is meshed with the sixth transmission gear 19 on the output shaft 4, forming a multi-stage transmission structure to meet the requirements of speed and torque under different working conditions, so that the reducer can adapt to a wider range of working scenarios.

Claims

1. A reducer maintenance and inspection system based on multi-source sensor fusion, including a gear reducer, data detection equipment, a data acquisition subsystem, a communication subsystem, a data analysis subsystem, and an intelligent diagnosis subsystem, characterized by: The gear reducer includes a housing, in which an input shaft, an output shaft, a high-speed shaft and a low-speed shaft are rotatably arranged, and the low-speed shaft, the high-speed shaft, the input shaft and the output shaft are connected by a gear set; The data detection equipment includes a three-axis vibration sensor, a microphone, a temperature sensor, a Hall speed sensor, a multi-parameter oil sensor, an adsorption-type wear particle sensor, and a water content sensor. The three-axis vibration sensor is respectively installed at the connection between the box body and the input shaft and the output shaft. The microphone is arranged on the outer periphery of the box body through an acoustic array to detect the internal noise of the gearbox, facilitate the location of faulty components and the analysis of fault problems. The installation position of the three-axis vibration sensor and the microphone forms a Cartesian coordinate system, and the Cartesian coordinate system is coplanar with the low-speed shaft, high-speed shaft, input shaft and output shaft. The temperature sensor is distributed on the high-speed shaft, low-speed shaft, input shaft and the connection between the output shaft and the box body. The Hall speed sensor is fixedly installed on the input shaft. The multi-parameter oil sensor, adsorption-type wear particle sensor, and water content sensor are arranged on the side of the box body away from the output shaft; The data acquisition subsystem is used to collect and integrate data from data detection equipment; The communication subsystem transmits the collected data to the data analysis system and the intelligent diagnosis system by wireless communication; The data analysis subsystem includes receiving data from the data acquisition subsystem and calculating the fault coordinates; The intelligent diagnosis subsystem includes a sub-database, and the intelligent diagnosis subsystem uses the fault coordinates calculated by the data analysis subsystem to perform fault determination on the component; The data analysis subsystem algorithm specifically includes the following steps: Step S1: Data preparation, receiving data from the data acquisition subsystem; Step S2: Determine the location coordinates (x, y) of the fault source by the time difference between the sound propagation times to different locations. The formula used is as follows: : represents the time it takes for the sound to propagate from the fault source S(x,y) to the origin; d1 refers to the distance between the microphone and the origin coordinate on the X axis; : d2 is the distance between the microphone and the origin coordinate on the Y axis; Among them, the speed of sound in air is vair=343m / s; Step S3: Estimate the time delay difference by using the generalized cross-correlation phase transform algorithm, and use the time difference to establish and solve the hyperbola equation to obtain the signal source; Step S4: Obtain the coordinates of the fault source through the Newton iteration method, and further determine the fault source location through particle filter fusion optimization, associate multiple database data with the fault source state, and then the fault source location changes over time. In the particle filter iteration process, initialization is first performed, and the generated particles obey , and then update the weights; Step S5: Resample to retain high-weight particles and perform weighted averaging to obtain the final coordinates: ； in, Refers to the weight of the i-th particle at time k.

2. The reducer maintenance and inspection system based on multi-source sensor fusion according to claim 1 is characterized in that: The three-axis vibration sensor and the microphone form a Cartesian coordinate system with the corresponding relationship being: the microphone is respectively installed on the X-axis, Y-axis and the coordinate origin, and the three-axis vibration sensor is respectively installed on the X-axis and Y-axis to make the coordinate plane coplanar with the gear shaft axis, and all fault sources are inside the coordinate system.

3. The reducer maintenance and inspection system based on multi-source sensor fusion according to claim 1 is characterized in that: The step S3 includes step S3.1: estimating the delay difference by using a generalized cross-correlation phase transformation algorithm, and correcting the delay difference. The formula used is as follows: ； Among them, Δtcomp is the corrected delay difference, dwal l is the box wall thickness, vwall=5000m / s is the speed of sound in the box wall, dair is the distance of the air part in the box, vair is the speed of sound in air; Step S3.2 uses the time difference to establish and solve the hyperbola equation to obtain the signal source. The hyperbola equation is: ； ； in, t1 is the time required for the signal to propagate from the target point to the reference point (d1,0), t0 is the time required for the signal to propagate from the target point to the origin (0,0), and Δt10 is the difference between these two propagation times; t2 is the time required for the signal to propagate from the target point to the reference point (0, d2), t0 is the time required for the signal to propagate from the target point to the origin (0, 0), and Δt20 is the difference between these two propagation times.

4. The reducer maintenance and inspection system based on multi-source sensor fusion according to claim 1 is characterized in that: The step S4 includes step S4.1: obtaining the coordinates of the fault source by the Newton iteration method, further determining the location of the fault source by particle filter fusion optimization, and associating multiple database data with the fault source status. The formula used is as follows: ； in is the state vector, containing Fault source coordinates at time and its rate of change , is the state transition matrix, is the process noise; Step S4.2: The subsequent fault source location changes over time and the observation equation is listed: ； in, is the observation vector, including RMS, which is the root mean square value of the vibration signal. is the bearing temperature rise rate, unit is ℃ / min, is the acoustic initial value coordinate, H is the observation matrix, and the state vector is mapped to the observation space. is the observation noise, which obeys the Gaussian distribution N(0,R), R is the noise covariance matrix; Step S4.3: During the particle filter iteration process, initialization is first performed and the generated particles obey , and then update the weights. The weight update formula is: ； is the actual observed value, Refers to the predicted observation value of the i-th particle.

5. The reducer maintenance and inspection system based on multi-source sensor fusion according to claim 1 is characterized in that: The intelligent diagnosis subsystem algorithm specifically includes the following steps: Step Q1: After obtaining the coordinates from the data analysis subsystem, the component is judged by first roughly comparing the samples and then loading the sub-database according to the coordinates; Step Q2: Use the dynamic time warping method to align the real-time features and samples and calculate the comprehensive similarity. The calculation formula is: ； in, is the real-time vibration eigenvector, is the database sample feature vector, Temperature feature similarity, Acoustic positioning coordinate matching: If S ≥ 0.95, the fault type is directly determined. If 0.7 ≤ S ≤ 0.9, the data is uploaded to the cloud, triggering secondary verification.

6. The reducer maintenance and inspection system based on multi-source sensor fusion according to claim 5 is characterized in that: The step Q1 includes step Q1.1: reducing the 32-dimensional vibration feature to 8 dimensions by PCA dimensionality reduction technology. The dimensionality reduction formula is: ； Among them, V is the original 32-dimensional feature vector, P is the principal component matrix; Step Q1.2: Use INT8 quantization technology to convert 32-bit floating-point numbers into 8-bit integers to reduce the amount of calculation and speed up inference. The quantization formula is: Here, μ is the data mean, σ is the data standard deviation, and FP32 floating-point numbers are compressed to INT8 integers, reducing the model memory usage from 32MB to 8MB and shortening the inference time.

7. The reducer maintenance and inspection system based on multi-source sensor fusion according to claim 6 is characterized in that: Step Q2 includes step Q2.1: The secondary verification interval is 0.7≤S≤0.

9. The system will immediately upload the relevant data to the cloud to further determine the type, severity and possible development trend of the fault through the cloud; Step Q2.1: The cloud feeds back the judgment results to the local device and provides corresponding maintenance suggestions based on the fault situation.

8. The reducer maintenance and inspection system based on multi-source sensor fusion according to any one of claims 1-2, characterized in that: The input shaft is arranged perpendicular to the high-speed shaft, and the high-speed shaft, the low-speed shaft and the output shaft are arranged in parallel.

9. The reducer maintenance and inspection system based on multi-source sensor fusion according to any one of claims 1-2, characterized in that: The gear set includes a first rotating gear arranged on the output shaft toward one end of the high-speed shaft, the high-speed shaft is provided with a second transmission gear meshing with the first transmission gear, the high-speed shaft is provided with a third transmission gear located on one side of the second transmission gear, the low-speed shaft is provided with a fourth transmission gear meshing with the third transmission gear, the low-speed shaft is provided with a fifth transmission gear located on one side of the fourth transmission gear, and the output shaft is provided with a sixth transmission gear meshing with the fifth transmission gear.

Images