Unmanned AGV power transmission system vibration diagnosis monitoring method, device and equipment

Abstract

The application discloses an unmanned AGV power transmission system vibration diagnosis monitoring method, device and equipment, and relates to the technical field of equipment state monitoring; a vibration monitoring device is used to collect state monitoring data of multiple drive transmission components of the unmanned AGV; a density-based clustering algorithm is used to identify the working conditions of the state monitoring data and filter out noise, so that other working condition data other than the noise and normal working conditions is filtered out, and normal working condition data is obtained; feature extraction is performed on the normal working condition data, multi-dimensional feature data extracted is input into a Gaussian mixture model to obtain a fault level early warning result, the multi-dimensional feature data is input into a fault diagnosis model based on a convolutional neural network to obtain an unmanned AGV fault type diagnosis result, and the results are visually displayed. The application can accurately capture weak and transient fault features, realize accurate fault grading early warning and fault diagnosis, and improve the operation reliability and operation efficiency of the unmanned AGV.

Description

Technical Field

[0001] This invention belongs to the field of equipment condition monitoring technology, and specifically relates to a vibration diagnosis and monitoring method, device and equipment for unmanned AGV power transmission system. Background Technology

[0002] With the accelerating pace of intelligent and digital transformation in manufacturing, production line environments are becoming increasingly complex and dynamic, placing higher demands on the autonomy, flexibility, and operational reliability of material handling equipment. Automated Guided Vehicles (AGVs), as the core carrier for flexible logistics, have become a key support for modern smart factories and logistics systems due to their autonomous navigation, precise positioning, and task execution capabilities. Their transport capacity is highly dependent on the stable and efficient operation of the power transmission system. The health of the power transmission system directly determines the overall operating efficiency, reliability, and service life of the vehicle. Damage to its core components, such as the drive motor, gearbox / reduction gearbox, drive shaft and coupling, and wheel bearings, can cause significant vibrations, leading to performance degradation or even sudden failure. Therefore, effective vibration monitoring of the power transmission system of unmanned AGVs is of paramount importance: it is the foundation for ensuring the continuity of core handling functions, and can prevent abnormal vehicle operation by detecting potential faults in components such as bearings, gears, and motors at an early stage. It can also reduce unplanned downtime through predictive maintenance, realizing the transformation from "reactive maintenance" to "predictive maintenance". At the same time, for unmanned AGVs performing precision handling, vibration monitoring can also indirectly ensure product quality and prevent cargo displacement or collisions caused by abnormal vibration.

[0003] However, vibration monitoring of the power transmission system of unmanned AGVs faces severe challenges due to their unique operating conditions, with the core issue being their complex dynamic characteristics. First, unmanned AGVs frequently start, stop, decelerate, maintain constant speed, and standby according to task commands. This intermittent operating mode and variable speed characteristics cause the power transmission system to experience drastic fluctuations in load and speed. Fault characteristic frequencies change with speed, making traditional threshold alarms prone to false alarms, increasing the difficulty of feature extraction and accurate fault type identification. Second, due to frequent changes in operating conditions, vibration signals exhibit strong non-stationarity and non-linearity, rendering general feature extraction methods (such as fixed-bandwidth analysis) ineffective. Furthermore, irrelevant vibration noise introduced by complex workshop environmental factors (such as bumpy workshop surfaces and vibrations from other equipment) masks true fault characteristics, further interfering with the identification of actual transmission system fault signals.

[0004] Currently, equipment fault diagnosis technology based on vibration signals has been extensively studied. For example, patent CN119046713A discloses a vibration early warning method for submersible electric pumps based on DBSCAN (Density-Based Spatial Clustering of Applications with Noise, or simply density-based clustering). This method can achieve anomaly detection. However, it directly uses the density-based clustering algorithm for the final alarm, failing to solve the problem of multi-condition identification for unmanned AGVs. It cannot distinguish between acceleration / deceleration and constant speed conditions, easily misjudging normal operating fluctuations of unmanned AGVs as faults, resulting in an extremely high false alarm rate. Another patent, CN110647830A, discloses a bearing fault diagnosis method based on convolutional neural networks and Gaussian mixture models. However, its feature extraction does not consider the characteristics of non-stationary signals in unmanned AGVs, and the model training is based on steady-state data, making it difficult to adapt to the dynamic changes in fault characteristic frequencies under variable speed conditions of unmanned AGVs. The diagnostic accuracy will significantly decrease in practical applications of unmanned AGVs. In addition, patents such as the analog circuit fault diagnosis method based on wavelet analysis and finite Gaussian mixture model EM method disclosed in CN103064009A, which are based on traditional signal processing methods, also suffer from the common defects of inaccurate feature extraction and poor generalization ability when facing strong background noise and non-stationary signals of unmanned AGVs.

[0005] In summary, existing fault diagnosis technologies are mostly developed for large rotating machinery operating at constant speed or in a steady state, and cannot effectively address the three major challenges unique to unmanned AGVs: "operating condition fluctuations," "signal non-stationarity," and "strong environmental noise." This makes it difficult for general methods to accurately separate the true fault characteristics from complex vibration signals when applied to unmanned AGVs, thus failing to achieve reliable early warning and accurate diagnosis.

[0006] Therefore, due to the decisive role of key components such as bearings, gears, motors, and gearboxes in the power transmission system on the reliability of unmanned AGVs, and the unique fault risks and diagnostic challenges they face under special working conditions such as intermittent start-stop, drastic load changes, and wide-range speed changes, it is urgent to develop a set of intelligent vibration diagnosis and monitoring methods and devices specifically for unmanned AGVs. Summary of the Invention

[0007] To address the aforementioned issues, this invention proposes a vibration diagnosis and monitoring method, device, and equipment for unmanned AGV power transmission systems. This method can accurately identify operating conditions and effectively reduce noise under varying operating conditions. It can also accurately capture weak and transient fault characteristics in environments with strong noise interference and non-stationary signals, thereby achieving precise fault classification and early warning for different fault levels and accurate fault diagnosis. This improves the reliability and efficiency of unmanned AGV operation.

[0008] In a first aspect, the present invention provides a vibration diagnosis and monitoring method for an unmanned AGV power transmission system, comprising:

[0009] Vibration monitoring devices are used to collect status monitoring data of multiple drive and transmission components of unmanned AGVs;

[0010] Density-based clustering algorithms are used to identify operating conditions and filter out noise from the condition monitoring data, so as to filter out noise and other operating condition data outside of normal operating conditions, and obtain the normal operating condition data corresponding to the normal driving conditions of the unmanned AGV.

[0011] Feature extraction is performed on the normal operating condition data to obtain multi-dimensional feature data;

[0012] The multi-dimensional feature data is input into a trained Gaussian mixture model to perform graded fault warning and obtain fault level warning results.

[0013] The multi-dimensional feature data is input into a trained fault diagnosis model based on a convolutional neural network to perform fault diagnosis and obtain fault type diagnosis results for the unmanned AGV.

[0014] The fault level warning results and the unmanned AGV fault type diagnosis results are sent to the vehicle control system and / or host computer for visualization.

[0015] In an optional embodiment, the plurality of drive transmission components include a drive motor, a first-stage spur gear reducer, an oil pump motor, a brake valve body, and a second-stage axle bevel gear reducer. The vibration monitoring device includes a data acquisition device, a communication device, and sensors connected in sequence. The sensors include a first vibration acceleration sensor installed on the drive motor to measure vertical vibration information at the drive end of the drive motor; a second vibration acceleration sensor installed on the first-stage spur gear reducer to measure vertical vibration information at the input shaft of the first-stage spur gear reducer; and a third vibration acceleration sensor installed on the first-stage spur gear reducer to measure horizontal vibration information at the output shaft of the first-stage spur gear reducer. The device includes a fourth vibration acceleration sensor installed on the oil pump motor to measure the vertical vibration information of the oil pump motor drive end, a fifth vibration acceleration sensor installed on the brake valve body to measure the vibration information on the brake valve body, a sixth vibration acceleration sensor installed on the second-stage axle bevel gear reducer to measure the horizontal vibration information of the input shaft of the second-stage axle bevel gear reducer, and a key phase sensor installed on the drive shaft at the vertical frame position; wherein, the first-stage spur gear reducer transmits power to the second-stage axle bevel gear reducer through the drive shaft, and the drive shaft has a protrusion fixed at a position corresponding to the sensing surface of the key phase sensor to trigger the key phase sensor.

[0016] In an optional implementation, the density-based clustering algorithm has a minimum number of points ranging from 3 to 6, and a neighborhood radius ranging from 0.5 to 0.7.

[0017] In an optional implementation, the multi-dimensional feature data includes vibration acceleration features, velocity features, spectrum features, power spectral density function features, center frequency features, frequency standard deviation features, frequency domain maximum energy features, demodulation spectrum features, intrinsic mode component features, and optimal time-frequency representation features.

[0018] In an optional implementation, the step of extracting features from the normal operating condition data to obtain multi-dimensional feature data includes:

[0019] Vibration acceleration features, velocity features, spectral features, and power spectral density features are extracted based on the normal operating condition data.

[0020] The demodulation spectrum features are extracted by demodulating the normal operating condition data.

[0021] Based on the normal operating condition data, the intrinsic mode component features are extracted using the variational mode decomposition algorithm, wherein the penalty factor is set in the range of 1000~4000, the number of modes is set in the range of 4~7 (integers), and the noise tolerance is set in the range of 0.001~0.05.

[0022] Based on the normal operating condition data, the optimal time-frequency representation features are extracted using the entropy matching synchronous squeezing transformation algorithm. In the entropy matching synchronous squeezing transformation algorithm, the db4 wavelet basis is used for continuous wavelet transformation, and the gradient descent method is used to optimize the squeezing parameters to minimize the entropy, thereby obtaining the optimal time-frequency representation features.

[0023] In an optional implementation, the step of inputting the multi-dimensional feature data into a trained convolutional neural network-based fault diagnosis model to perform fault diagnosis and obtain a fault type diagnosis result for the unmanned AGV includes:

[0024] Harmonic feature data is extracted based on the normal operating condition data, and the multi-dimensional feature data is reduced in dimension to obtain the multi-dimensional feature data after dimension reduction.

[0025] The harmonic feature data and the dimensionality-reduced multidimensional feature data are input into a trained fault diagnosis model based on a convolutional neural network for fault diagnosis to obtain the fault type diagnosis result of the unmanned AGV; wherein, the fault diagnosis model based on a convolutional neural network includes a first convolutional layer, a second convolutional layer, a pooling layer, a first fully connected layer and a second fully connected layer connected in sequence.

[0026] In an optional implementation, the Gaussian mixture model is trained using the expectation-maximization algorithm.

[0027] Secondly, the present invention provides a vibration diagnosis and monitoring device for an unmanned AGV power transmission system, comprising:

[0028] The acquisition module is used to collect status monitoring data of multiple drive and transmission components of the unmanned AGV using a vibration monitoring device;

[0029] The preprocessing module is used to perform working condition identification and noise filtering on the status monitoring data using a density-based clustering algorithm, so as to filter out noise and other working condition data outside of normal working conditions, and obtain the normal working condition data corresponding to the normal driving working conditions of the unmanned AGV.

[0030] The feature extraction module is used to extract features from the normal operating condition data to obtain multi-dimensional feature data;

[0031] The fault classification and early warning module is used to input the multi-dimensional feature data into the trained Gaussian mixture model to perform graded fault early warning and obtain fault level early warning results.

[0032] The fault diagnosis module is used to input the multi-dimensional feature data into the trained fault diagnosis model based on convolutional neural network to perform fault diagnosis and obtain the fault type diagnosis result of the unmanned AGV.

[0033] The visualization module is used to send the fault level warning results and the unmanned AGV fault type diagnosis results to the vehicle control system and / or host computer for visualization display.

[0034] Thirdly, the present invention provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method described in any of the foregoing embodiments.

[0035] Fourthly, the present invention provides a computer-readable medium having processor-executable non-volatile program code, the program code causing the processor to perform the method described in any of the foregoing embodiments.

[0036] The beneficial effects of the technical solution provided by the embodiments of the present invention are as follows: The vibration diagnosis and monitoring method, device, and equipment for the unmanned AGV power transmission system of the present invention firstly collects the state monitoring data of multiple drive transmission components of the unmanned AGV using a vibration monitoring device, providing reliable multi-source input for subsequent analysis; then, it uses a density-based clustering algorithm to identify the working condition and filter out noise from the state monitoring data, thereby filtering out noise and other working condition data outside of normal working conditions, and obtaining the normal working condition data corresponding to the normal driving condition of the unmanned AGV. This enables automatic identification of normal driving states such as acceleration and constant speed, and abnormal states such as road bumps, under various working conditions such as intermittent start-stop and variable speed operation of the unmanned AGV, avoiding misjudging normal working condition fluctuations as faults and significantly reducing the false alarm rate; it also removes environmental interference and abnormal working condition data, obtaining pure positive data. Normal operating condition data is used to improve the input quality of subsequent processing modules, effectively reducing false alarms caused by operating condition fluctuations and environmental interference. Then, multi-dimensional feature extraction is performed on the normal operating condition data to transform non-stationary signals into stable and usable feature vectors, enhancing the expressive power of fault information. Next, the multi-dimensional feature data is input into a trained Gaussian mixture model to achieve hierarchical fault warnings. Simultaneously, a fault diagnosis model based on convolutional neural networks accurately identifies the fault type of the unmanned AGV. Finally, the warning and diagnosis results are integrated into the vehicle control system and / or the host computer for visualization. Through the organic synergy of the above steps, this invention systematically solves the problems of false alarms, difficulty in extracting non-stationary signal features, and strong noise interference in the power transmission system of unmanned AGVs under intermittent start-stop and variable speed operation, significantly improving operational reliability, operating condition adaptability, and maintenance efficiency. Attached Figure Description

[0037] Figure 1 A flowchart illustrating the vibration diagnosis and monitoring method for the power transmission system of an unmanned AGV provided in an embodiment of the present invention;

[0038] Figure 2 The schematic diagram of the unmanned AGV provided in the embodiment of the present invention is intended to illustrate the structure of the key transmission components, sensors, and on-board data acquisition system of the unmanned AGV.

[0039] Figure 3 This is a schematic diagram of the connection between the sensor and the communication system provided in an embodiment of the present invention;

[0040] Figure 4 This is a schematic diagram illustrating the identification results of working condition identification using a density-based clustering algorithm, provided in an embodiment of the present invention.

[0041] Figure 5 A schematic diagram of the structure of a fault diagnosis model based on a convolutional neural network provided in an embodiment of the present invention;

[0042] Figure 6A schematic diagram of fault classification and early warning results based on Gaussian mixture model provided in an embodiment of the present invention;

[0043] Figure 7 A schematic diagram illustrating the parameter distribution, feature map, and diagnostic results visualization of a fault diagnosis model based on a convolutional neural network provided in an embodiment of the present invention.

[0044] Figure 8 A schematic diagram of the system principle of the vibration diagnosis and monitoring device for the power transmission system of an unmanned AGV provided in an embodiment of the present invention;

[0045] Figure 9 A schematic diagram of the system principle of an electronic device provided in an embodiment of the present invention.

[0046] In the diagram: 001-Drive motor; 002-First-stage spur gear reducer; 003-Oil pump motor; 004-Brake valve body; 005-Second-stage axle bevel gear reducer; 100-Data acquisition device; 101-Key phase signal acquisition and conditioning board; 102-Acceleration signal acquisition and conditioning board; 103-Board frame; 104-Main control board; 105-Power module; 107-485 interface; 200-Communication device; 201-CAN to 485 communication module; 202-Repeater; 310-Key phase sensor; 321-First vibration acceleration sensor; 322-Second vibration acceleration sensor; 323-Third vibration... Accelerometer sensor; 324-Fourth vibration acceleration sensor; 325-Fifth vibration acceleration sensor; 326-Sixth vibration acceleration sensor; 401-Explosion-proof box; 501-Unmanned AGV cab; 601-Universal joint coupling; 602-Drive shaft; 603-Wheel; 604-Steering valve; 605-Gear pump; 10-Acquisition module; 20-Preprocessing module; 30-Feature extraction module; 40-Fault classification and early warning module; 50-Fault diagnosis module; 60-Visualization module; 1000-Electronic equipment; 1001-Communication interface; 1002-Processor; 1003-Memory; 1004-Bus. Detailed Implementation

[0047] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0048] Currently, vibration monitoring is typically used for stationary rotating equipment such as pumps, centrifuges, and reciprocating machines. During operation, the speed and load of these devices usually do not change significantly or rapidly. Threshold alarms are commonly used, which involve setting a threshold based on national standards or empirical values. There are few solutions for installing vibration sensors on outdoor unmanned AGVs. In addition, vehicles face unique fault risks and diagnostic challenges under special operating conditions such as intermittent start-stop, drastic load changes, and wide-range speed changes. Threshold-based alarm modes may generate more alarms due to interference under these conditions.

[0049] Based on this, this embodiment proposes a vibration diagnosis and monitoring method for the power transmission system of an unmanned AGV. First, a vibration monitoring device and sensor scheme for the unmanned AGV is constructed. Then, raw vibration data of key components are collected according to preset parameters, and noise reduction is performed using a density-based clustering algorithm to identify normal operating conditions. Next, a trained Gaussian mixture model is used for hierarchical fault early warning, and a pre-trained convolutional neural network-based fault diagnosis model is used for fault diagnosis. Finally, server visualization deployment is performed (in specific implementation, scheduling system integration verification can also be performed). The method of this embodiment is described in detail below.

[0050] Reference Figure 1 The method in this embodiment includes the following steps S100 to S600.

[0051] Step S100: Use a vibration monitoring device to collect status monitoring data of multiple drive and transmission components of the unmanned AGV.

[0052] Here, vibration monitoring devices such as Figure 2 It includes a data acquisition device 100, a communication device 200, and sensors. Multiple drive transmission components include a drive motor 001, a first-stage spur gear reducer 002, an oil pump motor 003, a brake valve body 004, and a second-stage axle bevel gear reducer 005.

[0053] like Figure 2As shown, the sensors include a first vibration acceleration sensor 321 installed on the drive motor 001 to measure vertical vibration information at the drive end of the drive motor; a second vibration acceleration sensor 322 installed on the first-stage spur gear reducer 002 to measure vertical vibration information at the input shaft of the first-stage spur gear reducer; a third vibration acceleration sensor 323 installed on the first-stage spur gear reducer 002 to measure horizontal vibration information at the output shaft of the first-stage spur gear reducer; and a fourth vibration acceleration sensor 324 installed on the oil pump motor 003 to measure vertical vibration information at the drive end of the oil pump motor. The system includes a fifth vibration acceleration sensor 325 that measures vibration information on the brake valve body 004, a sixth vibration acceleration sensor 326 that measures horizontal vibration information on the input shaft of the second-stage axle bevel gear reducer 005, and a key phase sensor 310 that is installed on the drive shaft 602 at the vertical frame position. The first-stage spur gear reducer 002 transmits power to the second-stage axle bevel gear reducer 005 via the drive shaft 602. The drive shaft 602 has a protrusion that triggers the key phase sensor at a position corresponding to the sensing surface of the key phase sensor.

[0054] The explosion-proof box 401 is installed in the unmanned AGV cab 501 using a fastening method. Figure 3 This is a schematic diagram showing the connection between the data acquisition device 100 and the communication device 200. The data acquisition device 100 includes a key phase signal acquisition and conditioning board 101, an acceleration signal acquisition and conditioning board 102, a board frame 103, a main control board 104, and a power module 105. The communication device 200 includes a CAN-to-485 communication module 201 and a repeater 202. The data acquisition device 100 transmits data to the outside world through a network port. The data acquisition device 100 communicates with the communication device 200 through a 485 interface 107. The communication device 200 is placed inside an explosion-proof box 401.

[0055] The outdoor unmanned AGV's transmission path is as follows: drive motor 001 drives the first-stage spur gear reducer 002 via universal joint coupling 601. The first-stage spur gear reducer 002, through universal joint coupling 601, drives the drive shaft 602 to transmit power to the second-stage axle bevel gear reducer 005 at the front and rear axles. The bevel gear reducer at the axle performs a second-stage reduction, and then the second-stage axle bevel gear reducer 005 transmits power to the third-stage planetary reducer at the wheels, driving the wheels 603 to rotate and thus controlling the vehicle's movement. After the vehicle is powered on, the oil pump motor 003 runs continuously, maintaining a constant speed to keep the oil pressure constant. The valve group mainly consists of ordinary solenoid valves and proportional solenoid valves, a steering valve 604 to control the AGV's steering, and a brake valve body 004 to control the brakes.

[0056] This embodiment includes six vibration acceleration sensors. The first vibration acceleration sensor 321, which detects vibration information of the drive motor 001, is installed vertically at the drive end of the drive motor 001 to monitor the vibration of the drive motor 001 and any abnormalities in the drive motor 001 bearings. The second vibration acceleration sensor 322, which detects vibration information of the first-stage spur gear reducer 002, is installed vertically at the input shaft of the first-stage spur gear reducer 002. The third vibration acceleration sensor 323, which detects vibration information of the first-stage spur gear reducer 002, is installed horizontally at the output shaft of the first-stage spur gear reducer 002 to monitor the vibration of the first-stage spur gear reducer 002 and any abnormalities in the bearings or gears inside the first-stage spur gear reducer 002. The fourth vibration acceleration sensor 324, which detects vibration information of the oil pump motor 003, is installed vertically at the drive end of the oil pump motor 003 to monitor the vibration of the oil pump motor 003, any bearing abnormalities, and the operation of the gear pump 605. The fifth vibration acceleration sensor 325, which detects vibration information of the brake valve body, is installed on the brake valve body 004 to monitor the valve body's condition. The sixth vibration acceleration sensor 326, which detects vibration information of the second-stage axle bevel gear reducer, is installed horizontally on the input shaft of the second-stage axle bevel gear reducer 005 to monitor bearing and gear abnormalities in the second-stage axle bevel gear reducer 005. The sensor also includes a key phase sensor 310, which is fixed to the vehicle frame above the drive shaft 602 between the first-stage spur gear reducer 002 and the second-stage axle bevel gear reducer 005 using a mounting bracket. A protrusion is formed by welding iron sheets onto the vibration measuring band of the drive shaft 602 aligned with the key phase sensor. This protrusion triggers the key phase measuring point to obtain the real-time vehicle speed.

[0057] This embodiment uses six vibration acceleration sensors to collect data from the target unmanned AGV based on a preset sampling frequency and duration to obtain raw vibration data of each key component. Drive shaft speed data is collected using a key phase sensor. The status monitoring data includes raw vibration data and speed data. In this embodiment, other sensors (such as temperature sensors) can also be configured, requiring corresponding sensor configuration and signal acquisition and conditioning boards. Therefore, the status monitoring data also includes data from these other sensors.

[0058] Currently, unmanned AGVs have complex internal layouts and diverse components. The AGV transmission chain includes motors, multi-stage reduction gearboxes, drive shafts, brake valve bodies, etc., resulting in complex vibration propagation paths and requiring multi-point monitoring. This embodiment features a specially designed sensor layout. First, precise test point selection and orientation optimization were performed. Based on the dynamics analysis of the AGV transmission path, vibration acceleration sensors are precisely positioned as follows: vertically at the drive end of the drive motor to monitor motor bearing failures; vertically at the input shaft and horizontally at the output shaft of the first-stage spur gear reducer to detect gear meshing abnormalities and bearing failures; vertically at the drive end of the oil pump motor to monitor the oil pump's operating status; at the oil inlet of the brake valve body to detect abnormal valve body movement; and horizontally at the input shaft of the axle bevel gear reducer to monitor the second-stage reduction components. Second, special consideration was given to the installation of the key phase sensor. A protrusion is formed by welding iron sheets to trigger the key phase signal, accurately synchronizing speed and vibration data, providing a foundation for operational condition identification. The installation layout of these sensors can simultaneously capture characteristic vibrations from different components such as motors, gearboxes, bearings, and valve bodies, solving the problem of dispersed and difficult-to-locate fault sources in unmanned AGVs.

[0059] Step S200: Use density-based clustering algorithm to identify operating conditions and filter noise from the condition monitoring data, so as to filter out noise and other operating condition data outside the normal operating conditions, and obtain the normal operating condition data corresponding to the normal driving conditions of the unmanned AGV.

[0060] In an optional embodiment, in the density-based clustering algorithm, the minimum number of points ranges from 3 to 6, and the neighborhood radius ranges from 0.5 to 0.7. Preferably, in a specific implementation, the minimum number of points is 4, and the neighborhood radius is 0.6.

[0061] Specifically, this embodiment employs a density-based clustering method for operating condition identification. Based on the rotational speed information acquired by the vehicle's key phase sensors within a selected time period, the variation in vehicle rotational speed is obtained. The density-based clustering algorithm divides the key phase signals during vehicle operation, identifying the interval with the longest dwell time as the normal driving interval. This effectively filters out false alarms caused by low-frequency interference to the sensors due to road bumps or other reasons during acceleration, deceleration, and start-stop processes. The vibration signal from the normal driving interval is used for subsequent intelligent diagnostic analysis. The density-based clustering algorithm can discover clusters of arbitrary shapes and effectively handle noise points. It includes two key parameters: minimum number of points (MinPts) and neighborhood radius. A neighborhood radius that is too small will cause many points to be marked as noise, while a radius that is too large will cause multiple clusters to be merged. The minimum number of points, MinPts, refers to the minimum number of points required to form a cluster, and is usually chosen to be greater than or equal to the dimension + 1. The density-based clustering algorithm in this embodiment for working condition identification mainly includes the following steps (1) to (5).

[0062] (1) Initialization; select neighborhood radius Set the minimum number of points, MinPts, and mark all points as unvisited.

[0063] (2) Select a point; randomly select an unvisited point p and mark it as visited.

[0064] (3) Determine the core point; if p's If there are at least MinPts points in the neighborhood, a new cluster C is created, and p is called the core point; otherwise, p is marked as a noise point.

[0065] (4) Expand the cluster; and combine p and its clusters. Points within the neighborhood are added to the current cluster for p. For each point q in the neighborhood, if q has not been visited, it is marked as visited.

[0066] (5) Repeat steps (2) to (4) until all points have been visited.

[0067] In unmanned AGV (Automated Guided Vehicle) operating condition identification, the number of operating conditions and their fluctuations cannot be known in advance. Furthermore, the acceleration and deceleration of unmanned AGVs easily generate noise measurement points, and the dwell time varies in different operating conditions, resulting in different cluster densities. Due to these characteristics, density-based clustering algorithms are more suitable for identifying unmanned AGV operating conditions compared to other clustering algorithms. The results of clustering unmanned AGV operating conditions using a density-based clustering algorithm are as follows: Figure 4 As shown.

[0068] In practice, normal vehicle operation data is collected in advance during vehicle operation. Fault data, including motor bearing failure, reducer gear failure, rotating shaft bearing failure, misalignment, and loose rotating parts, is obtained through fault experiments and simulations. Based on the collected fault data, preprocessing is performed in step S200. A density-based clustering algorithm is used to preprocess each segment of raw data and fault data, filtering out interference signals generated by road bumps or other reasons during vehicle acceleration, deceleration, and start-stop processes. This yields normal operating range data that can be used for subsequent diagnostic model training; that is, normal operating condition data corresponding to the normal operating conditions of the unmanned AGV. Fault classification, early warning, and fault diagnosis are based on this data. Furthermore, the training of the Gaussian mixture model and the convolutional neural network-based fault diagnosis model is also based on the normal operating condition data obtained after preprocessing in step S200.

[0069] Currently, clustering algorithms are commonly used for customer grouping or image segmentation. This embodiment creatively applies a density-based clustering algorithm to unmanned AGVs. However, research has found that different operating conditions of unmanned AGVs (acceleration, constant speed, deceleration) will form clusters with different densities and shapes on the speed-time graph. Furthermore, the acceleration and deceleration processes will generate a large number of noise points (these data will affect model training and recognition accuracy when fed into the subsequent diagnostic model). The operating conditions of unmanned AGVs are unknown, the operating time is unpredictable, and the cluster shapes are irregular. Additionally, the acceleration and deceleration processes of unmanned AGVs are prone to generating noise points. This embodiment inputs the collected experimental data into a density-based clustering algorithm and optimizes the algorithm's core parameters (neighborhood radius ε, minimum number of points MinPts). Based on the analysis of data dimensions (rotation speed, vibration) and actual test data from unmanned AGV operation, the minimum number of points is controlled within the range of 3-6, and the neighborhood radius within the range of 0.5-0.7. The optimal parameter values ​​are MinPts=4 and neighborhood radius ε=0.6. This allows the density-based clustering algorithm to discover clusters of arbitrary shapes, corresponding to various operating conditions, while effectively identifying and removing noise data during acceleration and deceleration. This step is a crucial prerequisite for the effective application of the subsequent model. Without this processing, directly inputting non-stationary signals containing a large amount of noise into the model would lead to false alarms and diagnostic failures. This embodiment provides a specific and targeted treatment for the intermittent operation model of unmanned AGVs.

[0070] It is important to reiterate that the minimum number of points is defined based on the number of driving conditions. In this example, the minimum number of points, MinPts, is selected as an integer between 3 and 6. When it is less than 3, the algorithm becomes too sensitive to noise; when it is greater than 6, it is easy to miss some smaller cluster structures (corresponding to driving conditions with short dwell times but belonging to normal driving). The neighborhood radius ε is preferably between 0.5 and 0.7. Within this range, density-based clustering algorithms can best distinguish the inherent cluster structures in the data, avoiding over-segmentation or merging.

[0071] like Figure 4 As shown, after preprocessing in step S200, the data during unmanned AGV operation condition switching and the data under different operating conditions can be accurately identified (as shown in operating conditions 1 and 2 in the figure). After data preprocessing in step S200, feature extraction of the normal operating condition data is required in step S300 before the feature-extracted data can be input into the Gaussian mixture model and the fault diagnosis model based on convolutional neural network.

[0072] Step S300: Extract features from the normal operating condition data to obtain multi-dimensional feature data.

[0073] In optional embodiments, the multi-dimensional feature data includes one or more of the following: vibration acceleration features, velocity features, spectrum features, power spectral density function features, center frequency features, frequency standard deviation features, frequency domain maximum energy features, demodulation spectrum features, intrinsic mode component features, and optimal time-frequency representation features. Among these, vibration acceleration features, velocity features, spectrum features, power spectral density function features, center frequency features, frequency standard deviation features, and frequency domain maximum energy features are all conventional features.

[0074] In an optional embodiment, step S300 includes the following steps S301 to S304.

[0075] Step S301 involves extracting vibration acceleration features, velocity features, spectral features, and power spectral density features based on normal operating condition data. Vibration acceleration and velocity features are obtained from the normal operating condition data after data preprocessing in step S200. Spectral features are obtained by performing a Fourier transform on the normal operating condition data, decomposing it into single frequency components, which clearly reveals the main components of the signal. Power spectral density features are calculated from the spectrum, representing the power distribution with frequency. The center frequency feature, frequency standard deviation feature, and frequency domain maximum energy feature (reflecting energy changes in the spectrum) are specific characteristic parameters calculated from functions such as power spectral density.

[0076] Step S302: Demodulate and analyze the normal operating data to extract demodulation spectrum features. The demodulation spectrum features are obtained by bandpass filtering the normal operating data to focus on specific frequencies, then calculating the analytic signal using Hilbert transform to obtain the envelope signal, and finally performing a Fourier transform on the envelope signal. These demodulation spectrum features are used for early fault detection in rotating machinery such as rolling bearings and gears. This feature can effectively identify early, weak features in the gearboxes and bearings of unmanned AGVs.

[0077] Step S303: Based on normal operating condition data, intrinsic modal component features are extracted using the Variational Mode Decomposition (VMD) algorithm. The penalty factor is set within the range of 1000~4000, the number of modes is set within the range of 4~7 (integers), and the noise tolerance is set within the range of 0.001~0.05. The intrinsic modal component features and the optimal time-frequency representation features are also extracted from the normal operating condition data. The intrinsic modal component features are obtained by decomposing each intrinsic component using the Variational Mode Decomposition algorithm, which can adaptively obtain the frequency center and bandwidth of each mode and adaptively separate the modal components. Intrinsic modal component features have strong recognition capabilities for non-stationary and nonlinear signals of unmanned AGVs.

[0078] In this embodiment, when extracting intrinsic mode component features using the variational mode decomposition algorithm, the initial penalty factor is first set based on the actual unmanned AGV's feature information (rotation frequency: variable frequency, maximum motor operating speed 72Hz; gear meshing frequency: 1296Hz, 864Hz; bearing failure frequency: usually concentrated in high frequencies above 3000Hz) and adjusted and optimized using actual unmanned AGV data. =2000, initial mode number K=5 is used to extract target frequencies at each position to avoid mode aliasing. Considering that interference components are difficult to avoid during vehicle movement, a noise tolerance is selected. =0.01, while suppressing noise, the effective signal is preserved. The above adjustment can handle the non-stationary and nonlinear signals of unmanned AGVs and can adaptively separate the fault characteristic modes of different components.

[0079] In this embodiment, the penalty factor The value range is 1000~4000, which ensures sufficient accuracy in separating a wide range of frequency components from the rotational frequency (approximately 72Hz) to the high frequencies of bearing failure (above 3000Hz), especially effectively distinguishing close gear meshing frequencies (864Hz and 1296Hz). The initial mode number K ranges from 4 to 7, with a lower limit of 4 to cover the basic requirements of rotational frequency, gear meshing frequency, and bearing failure, while the upper limit of 7 provides tolerance for capturing other potential harmonics or weak features, avoiding over-decomposition. Noise tolerance The value ranges from 0.001 to 0.05 to accommodate the noise levels of unmanned AGVs under different operating conditions. Smaller values ​​can be selected for low-interference scenarios (such as smooth indoor surfaces), while larger values ​​can be selected for high-interference scenarios (such as unpaved outdoor surfaces) to enhance robustness against impulse noise.

[0080] Step S304: Based on normal operating condition data, the optimal time-frequency representation features are extracted using the entropy matching synchronous squeeze transform algorithm (SST). In the entropy matching synchronous squeeze transform algorithm, the db4 wavelet basis is used for continuous wavelet transform, and the gradient descent method is used to optimize the squeeze parameters to minimize the entropy, thereby obtaining the optimal time-frequency representation features.

[0081] Here, the optimal time-frequency representation feature is obtained by performing entropy matching synchronous squeezing transformation on the normal working condition data. This combines synchronous squeezing transformation and entropy optimization technology, introduces the entropy minimization criterion, and automatically adjusts the squeezing parameters. The specific steps are shown in steps (1) to (3) below.

[0082] (1) Initial time-frequency analysis

[0083] The initial time-frequency representation of the signal is calculated using either STFT (Short-Time Fourier Transform) or CWT (Continuous Wavelet Transform). .in These are parameters represented by time and frequency. t Indicates a time scale. Indicates the frequency scale.

[0084] (2) Synchrosqueezing Transform (SST)

[0085] Synchronous squeezing is applied to the initial time-frequency representation to redistribute energy:

[0086] (1)

[0087] In equation (1), It is the estimated instantaneous frequency. Represents the Dirac function, The parameters representing the time-frequency representation after compression. Represents the original parameters of the fundamental transformation. t Indicates a time scale. Indicates the frequency scale.

[0088] (3) Entropy calculation and optimization

[0089] Calculate the Shannon entropy in time-frequency representation EThis measures the degree of concentration in energy distribution.

[0090] (2)

[0091] In equation (2), It is a normalized time-frequency energy distribution.

[0092] By adjusting the squeezing parameters through optimization algorithms (such as gradient descent, particle swarm optimization, etc.), the entropy is minimized, thereby obtaining the optimal time-frequency representation.

[0093] In this embodiment, entropy matching synchronous extrusion transformation is used for unmanned AGV operation. Continuous wavelet transform is selected for time-frequency analysis (the wavelet basis function is db4, which can capture the high-frequency impact characteristics generated by bearings and gears, while suppressing slowly changing low-frequency noise). Gradient descent method is used to optimize and adjust the extrusion parameters to minimize entropy, thereby obtaining the optimal time-frequency representation. This allows for the clear extraction of the time and frequency components of the fault occurrence from transient signals under strong noise backgrounds of unmanned AGVs.

[0094] In the feature extraction stage, considering that the vibration signal of the unmanned AGV is non-stationary and non-Gaussian, simple time-domain features (such as RMS values) or frequency-domain features (such as the spectrum) will change under varying operating conditions, resulting in poor extraction results. This embodiment, for non-stationary signals, does not rely on a single feature, but constructs a feature combination of a multi-level, multi-domain feature system, including conventional features, demodulation spectrum features, intrinsic mode component features, and optimal time-frequency representation features, thereby enabling accurate feature extraction and laying a solid foundation for subsequent fault classification and diagnosis.

[0095] It should be noted that the feature extraction order of steps S302 to S304 is not limited to this. That is, step S303 or step S304 can be executed first to extract intrinsic mode component features or optimal time-frequency representation features.

[0096] Step S400: Input the multi-dimensional feature data into the trained Gaussian mixture model to perform graded fault warning and obtain the fault level warning result.

[0097] After obtaining multi-dimensional feature data using steps S100 to S300, a Gaussian mixture model is constructed and initialized. Random initialization is then performed. K The parameters of a Gaussian distribution, K The number of Gaussian distributions in the mixture model. k represent k A Gaussian distribution, with mean... Initialize using K-Means, covariance Initialized as an identity matrix, with mixing coefficients .

[0098] In an optional embodiment, the Gaussian mixture model is trained using the Expectation-Maximization Algorithm (EM algorithm), which is described in detail below.

[0099] Calculate the probability that a data point belongs to each cluster according to the following formula (3), and calculate the posterior probability (i.e., the probability of the data point belonging to each cluster). Belongs to the k The probability of a Gaussian distribution.

[0100] (3)

[0101] In formula (3) This indicates that the data points are within a given parameter. In the case of the first k The probability of being generated by a Gaussian distribution μ k Represents the mean vector. Represents the covariance matrix. Represents the mixing coefficient. Represents the given parameters No. k A Gaussian distribution; μ j Indicates the first j A vector of mean values ​​from a Gaussian distribution. j From 1 to K Positive integers.

[0102] The Gaussian distribution parameters are updated via the following steps (1) to (3) using the M-step (Maximization) method.

[0103] (1) Update the mean using the following formula (4).

[0104] (4)

[0105] In equation (4), This indicates that the data points are within a given parameter. In the case of the first k The probability of being generated by a Gaussian distribution μ k This represents the mean vector. i From 1 to N positive integers, N This represents the number of samples.

[0106] (2) Update the covariance using the following formula (5).

[0107] (5)

[0108] In equation (5), This indicates that the data points are within a given parameter. In the case of the first k The probability of being generated by a Gaussian distribution, μ k Represents the mean vector. Represents the covariance matrix. It is a vector The transpose of . i From 1 to N positive integers, N This represents the number of samples.

[0109] (3) Update the mixing coefficients using the following formula (6).

[0110] (6)

[0111] In formula (6) This indicates that the data points are within a given parameter. In the case of the first k The probability of being generated by a Gaussian distribution i From 1 to N positive integers, N This represents the number of samples.

[0112] After iterating the EM algorithm through the aforementioned steps, convergence is determined by the following formula (7), the log-likelihood function is calculated, and convergence is determined.

[0113] (7)

[0114] In equation (7), Represents a dataset, π Represents the mixing coefficient. μ Represents the mean vector. Σ Represents the covariance matrix. i From 1 to N positive integers, N Represents the number of samples; μ k Indicates the first k A vector of mean values ​​from a Gaussian distribution. Represents the covariance matrix. Represents the mixing coefficient. Represents the given parameters No. k A Gaussian distribution, This represents the log-likelihood function.

[0115] If the change in the likelihood function is less than the threshold, then the iteration stops.

[0116] Then anomaly detection is performed for new data points. xThe log-likelihood is calculated using the following formula (8).

[0117] (8)

[0118] In equation (8) Represents the mixing coefficient. Represents the given parameters No. k A Gaussian distribution, Represents new data points x The log-likelihood. k From 1 to K positive integers, K This represents the number of Gaussian distributions in the mixture model.

[0119] By selecting the log-likelihood of normal data, the first... Percentiles are used to identify outliers in the data.

[0120] (9)

[0121] In equation (9) The first digit represents the log-likelihood set of the normal training dataset. Percentile function.

[0122] according to Different values ​​determine whether the alarm is mild, moderate, or severe.

[0123] During training, the Gaussian mixture model extracts features from normal operating condition data after noise reduction and condition identification, and inputs these features into the model for training to obtain a Gaussian mixture model for the normal driving range. In application, simulated fault data is used to test and optimize the corresponding log-likelihood percentiles, obtaining various statistical indicators of normal data, the log-likelihood range of normal data, and the log-likelihood range of each warning level, thus achieving tiered warning systems.

[0124] This implementation uses only data from the normal operating range of the unmanned AGV to train the GMM and construct a "normal state" baseline. When an alarm is triggered, a graded mechanism is adopted, which sets different percentiles of the log-likelihood value to achieve three levels of early warning: mild, moderate, and severe, adapting to the risk levels of unmanned AGV operation.

[0125] Step S500: Input the multi-dimensional feature data into the trained fault diagnosis model based on convolutional neural network to perform fault diagnosis and obtain the fault type diagnosis result of the unmanned AGV.

[0126] In an optional embodiment, step S500 includes the following steps S501 and S502.

[0127] Step S501: Extract harmonic feature data based on normal operating condition data, and perform dimensionality reduction on the multi-dimensional feature data to obtain dimensionality-reduced multi-dimensional feature data.

[0128] Step S502: Input the harmonic feature data and the dimensionality-reduced multi-dimensional feature data into the trained fault diagnosis model based on convolutional neural network to perform fault diagnosis and obtain the fault type diagnosis result of the unmanned AGV; wherein, the fault diagnosis model based on convolutional neural network includes a first convolutional layer, a second convolutional layer, a pooling layer, a first fully connected layer and a second fully connected layer connected in sequence.

[0129] Here, the fault data collected in step S100 is preprocessed in step S200. Each segment of raw data and fault data is preprocessed using the working condition identification method to filter out interference signals caused by road bumps or other reasons during vehicle acceleration, deceleration, and start-stop. The normal driving range signal that can be used normally for subsequent diagnostic model training is obtained, i.e., normal working condition data. The fault diagnosis model based on convolutional neural network is trained based on the normal working condition data. For example, the steps for training the fault diagnosis model based on convolutional neural network include the following steps (1) to (3).

[0130] (1) Normal operating condition data is first processed by signal processing and feature extraction to obtain harmonic features. Harmonic features refer to the energy of the power frequency and harmonics such as second harmonic and third harmonic at the measuring point calculated by the key phase (speed) sensor in the spectrum generated after the original waveform (i.e., normal operating condition data) is transformed by Fourier transform.

[0131] (2) Construct a fault diagnosis model based on a convolutional neural network. The number of network layers and parameters of this model are adjusted based on the characteristics of unmanned AGV vibration data. It includes two convolutional layers (first and second convolutional layers), one pooling layer, and two fully connected layers (first and second fully connected layers). The model is as follows: Figure 5 As shown.

[0132] Figure 5The convolutional neural network structure shown includes a sequentially connected input layer, two convolutional layers, one pooling layer, and two fully connected layers. The input image size is 5×10, representing a single-channel grayscale image 10 pixels high and 5 pixels wide. The number "10" between the input layer and the first convolutional layer indicates that the first convolutional layer uses 10 independent convolutional kernels for feature extraction; the output feature map size of this convolutional layer is 4×10. The number "10" between the first and second convolutional layers indicates that the second convolutional layer also uses 10 convolutional kernels, and its output feature map size is 3×5. The network downsamples the output feature map of the second convolutional layer through a pooling layer to compress the data dimensionality and retain key features. Finally, the downsampled feature map is flattened and input into two fully connected layers (fc1 and fc2), where fc1 outputs a size of 10×1 and fc2 outputs a size of 2×1, further completing the classification decision and outputting the fault type diagnosis result of the unmanned AGV. The overall structure achieves end-to-end intelligent diagnosis from raw vibration signals to fault types through a combination of convolution, pooling, and full connection.

[0133] (3) The model is trained based on the Adam optimization algorithm.

[0134] During input, the multi-dimensional feature data extracted above is used, and after dimensionality reduction, it is combined with harmonic features as the input to the constructed fault diagnosis model based on convolutional neural network. The harmonic features correspond to the fault mechanism (such as different harmonic components corresponding to different fault modes). By combining mathematics and mechanism, the neural network model is trained using the Adam optimization algorithm to obtain the trained fault diagnosis model based on convolutional neural network.

[0135] In step S600, the fault level warning results and the fault type diagnosis results of the unmanned AGV are sent to the vehicle control system and / or the host computer for visualization.

[0136] Here, the computer program corresponding to the method of this embodiment is deployed to the host computer server. During vehicle operation, data, historical data, and case data can all be judged by this program. Before judgment, each piece of raw data and fault data is preprocessed by the working condition identification method in step S200 to filter out interference signals generated by road bumps or other reasons during vehicle acceleration, deceleration, and start-stop. The program only judges the filtered signals and gives the corresponding status. The status includes, but is not limited to, normal, motor bearing failure, reducer gear failure, rotating shaft bearing failure, misalignment failure, and loose rotating parts failure.

[0137] The current data is analyzed in the device status function of the host computer system, and the host computer (through the display module) provides a visual display. The host computer displays the fault diagnosis model, the diagnostic results of the current data, and suggested measures, such as... Figure 6The figure shows a schematic diagram of the fault classification and early warning results based on the Gaussian mixture model. The figure shows the fault type diagnosis results of the unmanned AGV, including motor imbalance, shaft misalignment, rotating shaft bearing failure, reducer gear failure, loose shaft components, and oil pump mechanical failure.

[0138] The host computer system analyzes the current acceleration data and displays the fault type diagnosis results of the vibration signal bearing using both the Gaussian mixture model and the convolutional neural network-based fault diagnosis model. For example, the current model parameter distribution histogram and diagnosis results of the convolutional neural network-based fault diagnosis model are shown below. Figure 7 As shown. Figure 7 The module displays the distribution maps of the weight parameters of the first convolutional layer, the bias parameters of the first convolutional layer, the weight parameters of the second convolutional layer, the bias parameters of the second convolutional layer, the feature map area (displaying feature extraction data), and the diagnostic conclusion area (the fault diagnosis results of the fault type of the unmanned AGV bearing based on the fault diagnosis model of the convolutional neural network). In addition, this display module can also view the vibration waveform, spectrum, demodulation spectrum, and various feature information.

[0139] The verification and deployment phases of this embodiment are designed closely around the real-world operating environment and maintenance needs of unmanned AGVs to ensure reliable diagnostic results and timely warnings. To verify the effectiveness and feasibility of this method, the method of this embodiment was developed into a program system application and integrated with the on-site unmanned AGV scheduling system. When a vehicle malfunctions, the system application can view the vehicle's vibration status and intelligent diagnostic conclusions. Simultaneously, if an anomaly occurs during vehicle operation, the vehicle can receive fault codes in real time.

[0140] This solution establishes a dedicated technical path for vibration diagnosis of unmanned AGVs, achieving targeted design of unmanned AGVs through five stages: perception layer, data preprocessing, feature extraction, diagnostic model, and system application.

[0141] For the sensing layer, this embodiment, based on the unmanned AGV transmission chain structure, installs sensor arrays in key parts such as the motor, gearbox input / output shaft, and brake valve body in specific directions such as vertical and horizontal to ensure that characteristic vibrations of each component are captured from the source.

[0142] For data preprocessing, this embodiment uses a density-based clustering algorithm to process rotation speed and vibration signals. By optimizing parameters, it effectively distinguishes between acceleration, constant speed, and deceleration conditions of unmanned AGVs and automatically filters out noise data during acceleration and deceleration to ensure the quality of data input into subsequent models.

[0143] For feature extraction, this embodiment comprehensively extracts the conventional spectrum of the vibration signal, the demodulation spectrum for bearings and gears, VMD components, and entropy matching synchronous extrusion transformation to characterize the non-stationary characteristics of unmanned AGVs under varying working conditions.

[0144] For the diagnostic model, this embodiment uses unmanned AGV operation data to train a Gaussian mixture model to achieve fault classification and early warning. The extracted features are dimensionality-reduced and combined with harmonic features as network input to construct a convolutional neural network model for fault classification. The two work together, providing early warning before diagnosis.

[0145] For system applications, this embodiment integrates the diagnostic results with the unmanned AGV scheduling system, which has a dedicated visualization interface that allows users to view vibration waveforms, spectra, demodulation spectra, various characteristic conditions, and intelligent diagnostic conclusions, forming a complete closed loop of monitoring, diagnosis, and operation and maintenance.

[0146] This embodiment can determine the vehicle's operating condition based on the correlation between vehicle operating speed and vibration. It can automatically identify the vehicle's operating condition based on the vehicle's operating conditions within a selected time period, thereby avoiding the influence of irrelevant vibration and noise introduced by environmental factors on the identification of actual transmission system fault signals.

[0147] In this embodiment, the vehicle's driving process is divided into at least three motion phases: an acceleration phase from a standstill to a constant speed, a constant speed phase where the speed remains constant (or substantially constant), and a braking phase from the start of deceleration to complete stop. This embodiment designs experiments to collect vibration data under different fault modes, considering vibration characteristics and spectral indices under different operating conditions. Combined with fault mechanisms, mechanistic and mathematical models of different faults are constructed. By comparing the characteristic parameters in real-time data during operation, the vibration state of the equipment is evaluated.

[0148] This embodiment proposes a method for vibration diagnosis, monitoring and condition assessment of key drive transmission components of unmanned AGV using at least six vibration sensors and one key phase sensor (which may be combined with other sensors), including the following steps (a) to (d).

[0149] (a) Continuously measure the vibration of key drive transmission components of unmanned AGV during vehicle operation using vibration sensors, and collect at least six acceleration sensor signals and one key phase sensor signal.

[0150] (b) The collected data is preprocessed to extract the characteristics of equipment fault-sensitive parameters, including vibration signal spectrum, power spectral density function, center frequency, frequency standard deviation, frequency domain maximum energy, vibration signal demodulation spectrum, intrinsic mode components based on VMD decomposition, and optimal time-frequency representation based on entropy matching synchronous squeezing transformation.

[0151] (c) Anomaly detection is performed by extracting the fault-sensitive parameter features of the equipment and the normal data features of the key drive transmission components of the unmanned AGV based on the Gaussian mixture model.

[0152] (d) When the current vibration data is detected to exceed the cluster space of the normal data, the corresponding position sensor of the key drive transmission component of the unmanned AGV issues an alarm. According to the degree of exceeding the limit, the alarm is defined as mild, moderate or severe.

[0153] In possible embodiments, the following steps (e) to (g) may also be performed.

[0154] (e) When an alarm is detected during the operation of a critical drive transmission component of an unmanned AGV, fault diagnosis can be performed through intelligent fault diagnosis based on a convolutional neural network.

[0155] (f) The vibration signal is first processed and feature extracted to obtain harmonic features, and then the harmonic features are used as input to the neural network.

[0156] (g) Diagnose bearing faults based on a two-dimensional convolutional neural network model, and output diagnostic status assessments and alarms of different levels.

[0157] In this embodiment, information such as vehicle speed, current, voltage, and torque can be imported from the vehicle control system via a communication device as an auxiliary basis for judgment. The data acquisition unit can communicate alarm information to the vehicle control system in real time.

[0158] In this embodiment, the results of monitoring, analyzing, diagnosing, and issuing early warnings based on data collected from key drive and transmission components of an unmanned AGV can be displayed on the system interface. A visualization layer is set up to present the collected data and model data. By visualizing the vibration data acquired by the data acquisition device and distributing it to the host computer system, this data can be displayed on the screen.

[0159] In this application, the embodiment involves both hardware and software implementation. The hardware consists of an onboard data acquisition system and a host server. The onboard data acquisition system includes a key phase sensor, a vibration acceleration sensor, a key phase signal acquisition and conditioning board, an acceleration signal acquisition and conditioning board, a board frame, and a main control board. The main control board includes a memory, a processor, and corresponding acquisition software. The host server includes a memory, a processor, a display, and corresponding data processing and analysis software. The onboard acquisition system receives data through an interface and transmits the data to the onboard system through this interface or other interfaces. Due to limitations in the data communication method, the onboard data acquisition system and the host server use data export and import software to copy the acquired data to the host server.

[0160] The software implementation of this embodiment can be accomplished through one or more specific applications. These applications perform relevant spectrum analysis, envelope analysis, and other methods based on data collected from the key drive and transmission components of the unmanned AGV to obtain fault characteristics of each vibration signal from complex vibration modes. Signal characteristics include: spectrum, power spectral density function, center frequency, frequency standard deviation, frequency domain maximum energy, vibration signal demodulation spectrum, intrinsic mode components based on VMD decomposition, and optimal time-frequency representation based on entropy-matched synchronous extrusion transform, etc.

[0161] In this embodiment, after signal extraction, the extracted fault feature vector can be compared with the feature vector of normal data at set values ​​to see if they fall within the same high-dimensional range. Based on the result, a current state assessment can be derived. The host computer server has intelligent diagnostic software that can analyze and diagnose the vehicle operation data collected by the data acquisition system, and supports automatically providing diagnostic conclusions and fault levels.

[0162] Regarding the possible data protocols used, the on-board data acquisition system and the vehicle control system communicate bidirectionally via CAN protocol to 485 protocol, and the host computer server supports multiple communication protocols.

[0163] Therefore, in the preferred embodiment, the vibration acceleration signal and its derived vibration velocity signal, as well as the vibration acceleration signal demodulation spectrum, the intrinsic mode components based on VMD decomposition, and the optimal time-frequency representation based on entropy matching synchronous extrusion transformation are selected. These parameters characterize the equipment operating status and monitor and diagnose the operating status of each drive and transmission component of the unmanned AGV.

[0164] In other words, this means that as long as the machine operates without faults, there is no imbalance, the wear parts have only slight wear and the bearings are undamaged, the measured vibration mode will be consistent with the normal data obtained through experimental mode, or consistent with the standard operating data stored as reference data or model data.

[0165] For example, when the damage area in a bearing of an unmanned AGV gradually expands, its composite vibration mode will change. Even before the unmanned AGV fails, the aforementioned analysis method can still be used to make a clear judgment about the key components of the unmanned AGV from the changes in the composite vibration mode: that is, changes in vibration components or amplitude changes and eigenvector changes at specific frequencies can be attributed to that component. In addition, by specifically evaluating the type, degree, and speed of changes in vibration components in the composite vibration mode, a judgment can be made about the current state of the unmanned AGV. This approach can not only detect early signs of component failure, but also pinpoint a developing or manifesting failure to a specific transmission component of the unmanned AGV.

[0166] The complete vibration diagnosis and monitoring method and system for the power transmission system of an unmanned AGV provided in this application embodiment first acquires vibration signals of key parts during vehicle operation through acceleration sensors installed on key components of the unmanned AGV transmission system, such as the drive motor 001 and the first-stage spur gear reducer 002, and key phase sensors installed above the transmission shaft 602. The acceleration signal acquisition and conditioning board in the main control module can perform preliminary filtering processing on the vibration signals. The working condition identification method in step S102 of this embodiment is used to filter the pre-processed signals to obtain the normal driving range signals that can be input into the subsequent diagnostic model. Based on the already trained and deployed models (Gaussian mixture model and fault diagnosis model based on convolutional neural network), the system performs graded early warning and fault diagnosis on the signals in the selected time interval and gives corresponding conclusions. The system can visualize the corresponding diagnostic results in the display module. At the same time, the system has the function of analyzing and displaying the vibration velocity trend, acceleration trend, and corresponding acceleration and velocity waveforms, as well as the corresponding spectrum of the waveforms.

[0167] The method in this embodiment first constructs a comprehensive vibration monitoring device and sensor installation scheme for key drive and transmission components based on the target unmanned AGV. Then, it collects data from the target unmanned AGV based on a preset sampling frequency and sampling duration to obtain the original vibration data of each key component. The original vibration data of each key component is processed for noise reduction using a density-based clustering algorithm. The driving conditions of the data within a set time period are identified. For the signals after the working conditions are identified, fault features in the time domain, frequency, and time-frequency domain are extracted, and a Gaussian mixture model is used for fault classification and early warning. For the signals after the working conditions are identified, fault features such as harmonic features are obtained, and a convolutional neural network is used for fault diagnosis. The method in this embodiment can be deployed on a server for visualization. The effectiveness and feasibility of this implementation method are verified by integration with the on-site unmanned AGV scheduling system, thereby improving the reliability of unmanned AGVs during operation.

[0168] See Figure 8This invention provides a vibration diagnosis and monitoring device for an unmanned AGV power transmission system, comprising an acquisition module 10, a preprocessing module 20, a feature extraction module 30, a fault classification and early warning module 40, a fault diagnosis module 50, and a visualization module 60. The acquisition module 10 collects status monitoring data of multiple drive transmission components of the unmanned AGV using a vibration monitoring device. The preprocessing module 20 uses a density-based clustering algorithm to identify operating conditions and filter noise from the status monitoring data, removing noise and other operating conditions outside of normal operating conditions to obtain normal operating condition data corresponding to the normal operating conditions of the unmanned AGV. The feature extraction module 30 extracts features from the normal operating condition data to obtain multi-dimensional feature data. The fault classification and early warning module 40 inputs the multi-dimensional feature data into a trained Gaussian mixture model for graded fault early warning, obtaining fault level warning results. The fault diagnosis module 50 inputs the multi-dimensional feature data into a trained convolutional neural network-based fault diagnosis model for fault diagnosis, obtaining fault type diagnosis results for the unmanned AGV. The visualization module 60 is used to send the fault level warning results and the fault type diagnosis results of the unmanned AGV to the vehicle control system and / or host computer for visualization display.

[0169] In an optional embodiment, the multiple drive transmission components include a drive motor, a first-stage spur gear reducer, an oil pump motor, a brake valve body, and a second-stage axle bevel gear reducer. The vibration monitoring device includes a data acquisition device, a communication device, and sensors connected in sequence. The sensors include a first vibration acceleration sensor installed on the drive motor to measure vertical vibration information at the drive end of the drive motor; a second vibration acceleration sensor installed on the first-stage spur gear reducer to measure vertical vibration information at the input shaft of the first-stage spur gear reducer; and a third vibration acceleration sensor installed on the first-stage spur gear reducer to measure horizontal vibration information at the output shaft of the first-stage spur gear reducer. The system includes a speed sensor, a fourth vibration acceleration sensor installed on the oil pump motor to measure vertical vibration information at the drive end of the oil pump motor, a fifth vibration acceleration sensor installed on the brake valve body to measure vibration information on the brake valve body, a sixth vibration acceleration sensor installed on the second-stage axle bevel gear reducer to measure horizontal vibration information on the input shaft of the second-stage axle bevel gear reducer, and a key phase sensor installed on the drive shaft at the vertical frame position. The first-stage spur gear reducer transmits power to the second-stage axle bevel gear reducer via the drive shaft, and the drive shaft has a protrusion that triggers the key phase sensor at a position corresponding to the sensing surface of the key phase sensor.

[0170] In an optional embodiment, in the density-based clustering algorithm, the minimum number of points ranges from 3 to 6, and the neighborhood radius ranges from 0.5 to 0.7.

[0171] In optional embodiments, the multi-dimensional feature data includes vibration acceleration features, velocity features, spectrum features, power spectral density function features, center frequency features, frequency standard deviation features, frequency domain maximum energy features, demodulation spectrum features, intrinsic mode component features, and optimal time-frequency representation features.

[0172] In an optional embodiment, the feature extraction module 30 includes a first extraction module, a second extraction module, a third extraction module, and a fourth extraction module. The first extraction module is used to extract vibration acceleration features, velocity features, spectral features, and power spectral density features based on normal operating condition data. The second extraction module is used to perform demodulation analysis on the normal operating condition data to extract demodulation spectral features. The third extraction module is used to extract intrinsic mode component features based on normal operating condition data using a variational mode decomposition algorithm, wherein the penalty factor is set within the range of 1000~4000, the number of modes is set within the range of 4~7 (integers), and the noise tolerance is set within the range of 0.001~0.05. The fourth extraction module is used to extract optimal time-frequency representation features based on normal operating condition data using an entropy-matched synchronous squeezing transform algorithm, wherein the entropy-matched synchronous squeezing transform algorithm uses a db4 wavelet basis for continuous wavelet transform and uses gradient descent to optimize the squeezing parameters to minimize entropy, thus obtaining the optimal time-frequency representation features.

[0173] In an optional embodiment, the fault diagnosis module 50 includes a feature dimensionality reduction module and a feature input module. The feature dimensionality reduction module is used to extract harmonic feature data based on normal operating condition data and reduce the dimensionality of the multi-dimensional feature data to obtain dimensionality-reduced multi-dimensional feature data. The feature input module is used to input the harmonic feature data and the dimensionality-reduced multi-dimensional feature data into a trained fault diagnosis model based on a convolutional neural network for fault diagnosis, and obtain the fault type diagnosis result of the unmanned AGV; wherein, the fault diagnosis model based on the convolutional neural network includes a first convolutional layer, a second convolutional layer, a pooling layer, a first fully connected layer, and a second fully connected layer connected in sequence.

[0174] In an optional embodiment, the Gaussian mixture model is trained using the expectation-maximization algorithm.

[0175] The apparatus provided in the embodiments of this application has the same inventive concept as the method provided in the embodiments of this application. As long as the method can solve the technical problem, the apparatus can also solve the technical problem. This will not be elaborated here.

[0176] Reference Figure 9The present invention also provides an electronic device 1000, including a communication interface 1001, a processor 1002, a memory 1003, and a bus 1004. The processor 1002, the communication interface 1001, and the memory 1003 are connected via the bus 1004. The memory 1003 is used to store a computer program that supports the processor 1002 in executing the vibration diagnosis and monitoring method of the unmanned AGV power transmission system. The processor 1002 is configured to execute the program stored in the memory 1003.

[0177] Optionally, embodiments of the present invention also provide a computer-readable medium having non-volatile program code executable by a processor 1002, the program code causing the processor 1002 to execute the vibration diagnosis and monitoring method for the unmanned AGV power transmission system as described in the above embodiments.

[0178] Those skilled in the art will understand that all or part of the steps of the methods described in the foregoing embodiments can be implemented by a computer program controlling related hardware. This computer program can be stored in a non-volatile computer-readable storage medium and, when executed, includes the steps of the foregoing methods.

[0179] It should be noted that the memory, storage, database, or other storage medium mentioned in the embodiments of this application all include at least one of non-volatile memory and volatile memory. Specifically:

[0180] Examples of non-volatile memory include: read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical storage, etc.

[0181] Examples of volatile memory include random access memory (RAM) or external cache memory. RAM itself also comes in various forms, such as static RAM (SRAM) or dynamic RAM (DRAM).

[0182] Furthermore, the technical features in the above embodiments can be combined arbitrarily. For the sake of brevity, the specification does not exhaust all possible combinations. However, it should be understood that as long as these combinations do not contradict each other, they should all be considered within the scope of this specification.

[0183] As is known from common technical knowledge, this invention can be implemented through other embodiments that do not depart from its spirit or essential characteristics. Therefore, the disclosed embodiments described above are merely illustrative in all respects and are not the only ones. All modifications within the scope of this invention or its equivalents are included in this invention.

Claims

1. A vibration diagnosis and monitoring method for an unmanned AGV power transmission system, characterized in that, include: Vibration monitoring devices are used to collect status monitoring data of multiple drive and transmission components of unmanned AGVs; Density-based clustering algorithms are used to identify operating conditions and filter out noise from the condition monitoring data, so as to filter out noise and other operating condition data outside of normal operating conditions, and obtain the normal operating condition data corresponding to the normal driving conditions of the unmanned AGV. Feature extraction is performed on the normal operating condition data to obtain multi-dimensional feature data; The multi-dimensional feature data is input into a trained Gaussian mixture model to perform graded fault warning and obtain fault level warning results. The multi-dimensional feature data is input into a trained fault diagnosis model based on a convolutional neural network to perform fault diagnosis and obtain fault type diagnosis results for the unmanned AGV. The fault level warning results and the unmanned AGV fault type diagnosis results are sent to the vehicle control system and / or host computer for visualization. The multi-dimensional feature data includes vibration acceleration features, velocity features, spectral features, power spectral density function features, center frequency features, frequency standard deviation features, frequency domain maximum energy features, demodulation spectrum features, intrinsic mode component features, and optimal time-frequency representation features; the feature extraction of the normal operating condition data to obtain multi-dimensional feature data includes: Based on the normal operating condition data, vibration acceleration features, velocity features, spectral features, and power spectral density function features are extracted; The demodulation spectrum features are extracted by demodulating the normal operating condition data. Based on the normal operating condition data, the intrinsic mode component features are extracted using the variational mode decomposition algorithm, wherein the penalty factor is set in the range of 1000~4000, the number of modes is set in the range of 4~7 (integers), and the noise tolerance is set in the range of 0.001~0.

05. Based on the normal operating condition data, the optimal time-frequency representation features are extracted using the entropy matching synchronous squeezing transformation algorithm. In the entropy matching synchronous squeezing transformation algorithm, the db4 wavelet basis is used for continuous wavelet transformation, and the gradient descent method is used to optimize the squeezing parameters to minimize the entropy, thereby obtaining the optimal time-frequency representation features.

2. The vibration diagnosis and monitoring method for the power transmission system of an unmanned AGV according to claim 1, characterized in that, The plurality of drive transmission components include a drive motor, a first-stage spur gear reducer, an oil pump motor, a brake valve body, and a second-stage axle bevel gear reducer. The vibration monitoring device includes a data acquisition device, a communication device, and sensors connected in sequence. The sensors include a first vibration acceleration sensor installed on the drive motor to measure vertical vibration information at the drive end of the drive motor; a second vibration acceleration sensor installed on the first-stage spur gear reducer to measure vertical vibration information at the input shaft of the first-stage spur gear reducer; a third vibration acceleration sensor installed on the first-stage spur gear reducer to measure horizontal vibration information at the output shaft of the first-stage spur gear reducer; a fourth vibration acceleration sensor installed on the oil pump motor to measure vertical vibration information at the drive end of the oil pump motor; a fifth vibration acceleration sensor installed on the brake valve body to measure vibration information on the brake valve body; a sixth vibration acceleration sensor installed on the second-stage axle bevel gear reducer to measure horizontal vibration information at the input shaft of the second-stage axle bevel gear reducer; and a key phase sensor installed on the drive shaft at a vertical position on the vehicle frame. The first-stage spur gear reducer transmits power to the second-stage axle bevel gear reducer via the drive shaft, and the drive shaft has a protrusion fixed at a position corresponding to the sensing surface of the key phase sensor to trigger the key phase sensor.

3. The vibration diagnosis and monitoring method for the power transmission system of an unmanned AGV according to claim 1, characterized in that, In the density-based clustering algorithm, the minimum number of points ranges from 3 to 6, and the neighborhood radius ranges from 0.5 to 0.

7.

4. The vibration diagnosis and monitoring method for the power transmission system of an unmanned AGV according to claim 1, characterized in that, The step of inputting the multi-dimensional feature data into a trained convolutional neural network-based fault diagnosis model to perform fault diagnosis and obtain fault type diagnosis results for the unmanned AGV includes: Harmonic feature data is extracted based on the normal operating condition data, and the multi-dimensional feature data is reduced in dimension to obtain the multi-dimensional feature data after dimension reduction. The harmonic feature data and the dimensionality-reduced multidimensional feature data are input into a trained fault diagnosis model based on a convolutional neural network for fault diagnosis to obtain the fault type diagnosis result of the unmanned AGV; wherein, the fault diagnosis model based on a convolutional neural network includes a first convolutional layer, a second convolutional layer, a pooling layer, a first fully connected layer and a second fully connected layer connected in sequence.

5. The vibration diagnosis and monitoring method for the power transmission system of an unmanned AGV according to claim 1, characterized in that, The Gaussian mixture model is trained using the expectation-maximization algorithm.

6. A vibration diagnosis and monitoring device for an unmanned AGV power transmission system, characterized in that, include: The acquisition module is used to collect status monitoring data of multiple drive and transmission components of the unmanned AGV using a vibration monitoring device; The preprocessing module is used to perform working condition identification and noise filtering on the status monitoring data using a density-based clustering algorithm, so as to filter out noise and other working condition data outside of normal working conditions, and obtain the normal working condition data corresponding to the normal driving working conditions of the unmanned AGV. The feature extraction module is used to extract features from the normal operating condition data to obtain multi-dimensional feature data; The fault classification and early warning module is used to input the multi-dimensional feature data into the trained Gaussian mixture model to perform graded fault early warning and obtain fault level early warning results. The fault diagnosis module is used to input the multi-dimensional feature data into the trained fault diagnosis model based on convolutional neural network to perform fault diagnosis and obtain the fault type diagnosis result of the unmanned AGV. The visualization module is used to send the fault level warning results and the unmanned AGV fault type diagnosis results to the vehicle control system and / or host computer for visualization display. Multidimensional feature data includes vibration acceleration features, velocity features, spectrum features, power spectral density function features, center frequency features, frequency standard deviation features, frequency domain maximum energy features, demodulation spectrum features, intrinsic mode component features, and optimal time-frequency representation features; The feature extraction module includes a first extraction module, a second extraction module, a third extraction module, and a fourth extraction module; the first extraction module is used to extract vibration acceleration features, velocity features, spectral features, and power spectral density function features based on normal operating condition data; The second extraction module is used to demodulate and analyze normal operating data to extract demodulation spectrum features; The third extraction module is used to extract intrinsic mode component features based on normal operating condition data and using variational mode decomposition algorithm. The penalty factor is set in the range of 1000~4000, the number of modes is set in the range of 4~7 integers, and the noise tolerance is set in the range of 0.001~0.

05. The fourth extraction module is used to extract the optimal time-frequency representation features based on normal operating condition data and using entropy matching synchronous squeezing transform algorithm. The entropy matching synchronous squeezing transform algorithm uses db4 wavelet basis for continuous wavelet transform and uses gradient descent method to optimize squeezing parameters to minimize entropy and obtain the optimal time-frequency representation features.

7. An electronic device, characterized in that, The method includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method according to any one of claims 1 to 5.

Images