Industrial robot fault diagnosis method, device and equipment and computer readable medium

Abstract

The invention relates to an industrial robot fault diagnosis method, device and equipment and a computer readable medium. The method comprises the steps that multiple pieces of operation state data of the industrial robot are acquired; preprocessing the operation state data to obtain enhanced data; performing feature extraction on the enhanced data, and constructing an operation feature vector of the industrial robot based on the extracted features; calculating the evidence support degree of each feature subset for different faults in the operation feature vector; and fusing all the evidence support degrees to obtain a fusion support degree, and determining that the industrial robot has a fault under the condition that the fusion support degree is greater than a preset alarm threshold value. According to the method, the multi-source operation state data is collected, preprocessing, feature vector construction and multi-feature subset evidence support degree calculation and fusion are performed, the multi-dimensional operation state of the robot is covered, the risk of misjudgment of a single feature is reduced, accurate fault recognition is achieved, and the technical problem that the accuracy of robot fault prediction is low is solved.

Description

Technical Field

[0001] This application relates to the field of industrial robot technology, and in particular to a method, apparatus, equipment and computer-readable medium for diagnosing industrial robot faults. Background Technology

[0002] In the field of industrial robot applications, production line downtime losses due to malfunctions are exorbitant, exceeding $20,000 per hour, and the repair costs for sudden malfunctions are 3-5 times higher than preventative maintenance. These malfunctions mainly stem from mechanical component failures, electrical faults, and control system anomalies. Traditional maintenance and monitoring methods are ineffective in addressing these issues, with the core problem being the low accuracy of robot malfunction prediction.

[0003] There is currently no effective solution to the problem of low accuracy in predicting robot faults. Summary of the Invention

[0004] This application provides a method, apparatus, equipment, and computer-readable medium for diagnosing industrial robot faults, in order to solve the technical problem of low accuracy in predicting robot faults.

[0005] According to one aspect of the embodiments of this application, this application provides a method for diagnosing industrial robot faults, including: acquiring multiple operating status data of the industrial robot; preprocessing the operating status data to obtain enhanced data; extracting features from the enhanced data and constructing an operating feature vector of the industrial robot based on the extracted features; calculating the evidence support degree of each feature subset in the operating feature vector for different faults; fusing all evidence support degrees to obtain fused support degree, and determining that the industrial robot has malfunctioned when the fused support degree is greater than a preset alarm threshold.

[0006] Optionally, acquiring multiple operating status data of the industrial robot includes at least one of the following: sending a vibration sampling command to a vibration sensor installed at the input / output flange of the harmonic reducer of the industrial robot at a first moment to obtain a vibration signal with a sampling timestamp as the target moment; sending a temperature sampling command to a temperature sensor embedded in the motor windings and / or harmonic reducer of the industrial robot at a second moment to obtain a temperature signal with a sampling timestamp as the target moment; sending a current sampling command to a current transformer clamped to the three-phase input line of the servo driver of the industrial robot at a third moment to obtain a current signal with a sampling timestamp as the target moment; and sending a motion sampling command to a motion sensor installed at the joint of the industrial robot at a fourth moment to obtain a motion signal with a sampling timestamp as the target moment; wherein the first, second, third, and fourth moments are used to ensure that different sensors all acquire signals according to the same sampling timestamp.

[0007] Optionally, before acquiring multiple operating status data of the industrial robot, the method further includes determining the command issuance time of each sensor in the following manner: connecting a common test signal to the input of all sampling channels and simultaneously issuing sampling commands to the sensors of all sampling channels; acquiring the capture timestamp of the common test signal captured by the sensors of each sampling channel; calculating the difference between each capture timestamp and a preset reference time as the sampling compensation for each sensor; acquiring the distance from each sensor to the processor and dividing the distance by the signal transmission speed to obtain the distance delay of each sensor; adding the preset reference time to the sampling compensation and distance delay of any sensor to obtain the command issuance time of the sensor.

[0008] Optionally, the operating status data is preprocessed to obtain enhanced data including at least one of the following: calling the target wavelet basis function to perform wavelet packet decomposition on the vibration signal according to a preset decomposition level to obtain the coefficient sequence of the first characteristic frequency band and the coefficient sequence of the second characteristic frequency band, wherein the first characteristic frequency band and the second characteristic frequency band are the fault-sensitive frequency bands of the harmonic reducer, and the enhanced data includes the coefficient sequence; calculating the spatial temperature gradient and the temporal temperature gradient of the temperature signal, wherein the enhanced data includes the spatial temperature gradient and the temporal temperature gradient; extracting current harmonics from the current signal, wherein the enhanced data includes the current harmonics.

[0009] Optionally, feature extraction of the enhanced data includes at least one of the following: calculating the energy entropy of the first characteristic frequency band based on the coefficient sequence of the first characteristic frequency band; calculating the energy entropy of the second characteristic frequency band based on the coefficient sequence of the second characteristic frequency band; calculating the vibration kurtosis of the vibration signal based on the mean, standard deviation, number of sampling points, and vibration acceleration of the vibration signal; calculating the ratio of the peak value to the effective value of the waveform as the peak factor based on the vibration waveform of the vibration signal; determining the maximum spatial temperature gradient based on the spatial temperature gradient; calculating the average temporal temperature gradient based on the temporal temperature gradient; calculating the maximum value of the second derivative of the temperature signal with respect to time to obtain the peak value of the temperature rise acceleration; and calculating the amplitude proportion of the 13th harmonic current based on the current harmonics. For current harmonics, calculate the total harmonic distortion rate; based on current harmonics, determine the content of negative sequence components in the current; calculate the absolute value of the robot joint output torque in the motion signal; calculate the absolute value of the robot joint angular velocity in the motion signal; where, the energy entropy of the first characteristic frequency band, the energy entropy of the second characteristic frequency band, the vibration kurtosis, and the peak factor are the vibration characteristics of the industrial robot; the maximum spatial temperature gradient, the average time temperature gradient, and the peak value of the temperature rise acceleration are the temperature characteristics of the industrial robot; the amplitude ratio of the 13th harmonic current, the total harmonic distortion rate, and the content of negative sequence components are the current characteristics of the industrial robot; the absolute value of the robot joint output torque and the absolute value of the robot joint angular velocity are the motion characteristics of the industrial robot.

[0010] Optionally, after determining that the industrial robot has failed, the method further includes: inputting at least one of vibration characteristics, temperature characteristics, current characteristics, and motion characteristics into a target physical model to predict the sustainable life of the industrial robot under failure conditions using the target physical model.

[0011] Optionally, before using the target physical model to predict the remaining lifetime, the method further includes obtaining the target physical model in the following manner: constructing an initial physical model; determining at least one of the wear increment, temperature gradient, vibration entropy value, and current distortion rate of the industrial robot as residual input terms based on operating state data; inputting the residual input terms into a long short-term memory neural network model and obtaining the residual correction terms output by the long short-term memory neural network model; and using the residual correction terms to correct the initial physical model to obtain the target physical model.

[0012] According to another aspect of the embodiments of this application, this application provides an industrial robot fault diagnosis device, including: an acquisition module for acquiring multiple operating status data of an industrial robot; a preprocessing module for preprocessing the operating status data to obtain enhanced data; a construction module for extracting features from the enhanced data and constructing an operating feature vector of the industrial robot based on the extracted features; a calculation module for calculating the evidence support degree of each feature subset in the operating feature vector for different faults; and a fusion decision module for fusing all evidence support degrees to obtain fused support degree, and determining that the industrial robot has malfunctioned when the fused support degree is greater than a preset alarm threshold.

[0013] According to another aspect of the embodiments of this application, this application provides an electronic device, including a memory, a processor, a communication interface and a communication bus. The memory stores a computer program that can run on the processor. The memory and the processor communicate with each other through the communication bus and the communication interface. When the processor executes the computer program, it implements the steps of the above method.

[0014] According to another aspect of the embodiments of this application, this application also provides a computer-readable medium having processor-executable non-volatile program code that causes the processor to perform the above-described method.

[0015] Compared with related technologies, the technical solutions provided in this application have the following advantages: This application provides a method for fault diagnosis of industrial robots, including: acquiring multiple operating status data of the industrial robot; preprocessing the operating status data to obtain enhanced data; extracting features from the enhanced data and constructing an operating feature vector of the industrial robot based on the extracted features; calculating the evidence support degree of each feature subset in the operating feature vector for different faults; fusing all evidence support degrees to obtain fused support degree; and determining that the industrial robot has malfunctioned when the fused support degree is greater than a preset alarm threshold. This application, by collecting multi-source operating status data, preprocessing it, constructing feature vectors, calculating and fusing evidence support degrees of multiple feature subsets, covers multi-dimensional operating states of the robot and reduces the risk of misjudgment based on a single feature, achieving accurate fault identification and solving the technical problem of low accuracy in robot fault prediction. Attached Figure Description

[0016] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0017] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, those skilled in the art can obtain other drawings based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of the hardware environment for an optional industrial robot fault diagnosis method provided according to an embodiment of this application; Figure 2 This is a schematic diagram of an optional industrial robot fault diagnosis method provided according to an embodiment of this application; Figure 3 This is a schematic diagram illustrating an optional physical model data-driven correction according to an embodiment of this application; Figure 4 This is a schematic diagram of the system architecture of an optional industrial robot fault diagnosis method provided according to an embodiment of this application; Figure 5 This is a block diagram of an optional industrial robot fault diagnosis device according to an embodiment of this application; Figure 6 This is a schematic diagram of an optional electronic device structure provided in an embodiment of this application. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] In the following description, the use of suffixes such as "module," "part," or "unit" to denote elements is solely for the purpose of illustration and has no specific meaning in itself. Therefore, "module" and "part" may be used interchangeably.

[0021] In the field of industrial robot applications, production line downtime losses due to malfunctions are exorbitant, exceeding $20,000 per hour, and the repair costs for sudden malfunctions are 3-5 times higher than preventative maintenance. These malfunctions mainly stem from mechanical component failures (accounting for 65%, such as reducer wear and bearing cracks), electrical faults (accounting for 25%, such as motor overheating and cable breakage), and control system anomalies (accounting for 10%, such as encoder drift). Traditional maintenance and monitoring methods are insufficient to effectively address these issues. The core problem lies in the inability to simultaneously achieve comprehensive perception of the operational status of key robot components, millisecond-level fault response, and high-precision lifespan prediction, resulting in low accuracy in robot fault prediction. Specifically, traditional reactive maintenance requires handling problems passively after they occur, leading to uncontrollable production interruptions. Regular maintenance, due to its rigid cycles, not only wastes 30% of resources but may also miss maintenance opportunities because it doesn't cover the accumulation period of hidden faults. Even if basic condition monitoring technology can trigger alarms, it has significant limitations—at the sensing level, external accelerometers are susceptible to environmental interference; for example, KUKA welding robots have a sensor failure rate as high as 15% in high-temperature environments, making it difficult to capture wear conditions in critical areas such as the joints. At the algorithm level, fault classification algorithms relying on the cloud, such as Siemens' cloud-based SVM, are affected by network latency, resulting in response times exceeding 2 seconds, failing to meet safety standards. At the model level, lifespan prediction methods based purely on historical statistical data, such as FANUC's related technologies, have prediction errors exceeding 30% for new collaborative robots, failing to provide a reliable basis for proactive maintenance.

[0022] To address the problems mentioned in the background art, according to one aspect of the embodiments of this application, an embodiment of an industrial robot fault diagnosis method is provided.

[0023] Optionally, in the embodiments of this application, the above-described industrial robot fault diagnosis method can be applied to, for example... Figure 1 The hardware environment shown consists of terminal 101 and server 103. Figure 1As shown, server 103 is connected to terminal 101 via a network and can be used to provide services to the terminal or clients installed on the terminal. Database 105 can be set up on the server or independently of the server to provide data storage services for server 103. The network mentioned above includes, but is not limited to, wide area network, metropolitan area network or local area network. Terminal 101 includes, but is not limited to, industrial robots, industrial robot control systems, PCs, mobile phones, tablets, etc.

[0024] An industrial robot fault diagnosis method in this application embodiment can be executed by server 103, or it can be jointly executed by server 103 and terminal 101, such as... Figure 2 As shown, the method may include the following steps: Step S202: Acquire multiple operating status data of the industrial robot; Step S204: Preprocess the running status data to obtain enhanced data; Step S206: Extract features from the augmented data and construct the operational feature vector of the industrial robot based on the extracted features; Step S208: Calculate the evidence support for different faults for each feature subset in the running feature vector; Step S210: Fuse all evidence support scores to obtain fused support score, and determine that the industrial robot has malfunctioned if the fused support score is greater than a preset alarm threshold.

[0025] In this embodiment, the multiple operating status data of the industrial robot refers to multi-source heterogeneous data that can reflect the mechanical, electrical, and motion states of the industrial robot during operation, including but not limited to vibration signals, temperature signals, current signals, and motion signals of key robot components. The aforementioned preprocessing refers to the process of performing signal enhancement, interference removal, and feature frequency band extraction on the collected raw operating status data, aiming to improve data quality and highlight fault-related feature information. Evidence support refers to the degree of support of each feature subset in the operating feature vector for different fault hypotheses such as normal operation, reducer wear, motor overheating, and motion deviation, based on DS evidence theory. The value range is [0,1], with a larger value indicating stronger support for the corresponding fault hypothesis. Fusion support refers to the result obtained by comprehensively calculating the evidence support of each feature subset through the orthogonal sum rule of DS evidence theory, used to eliminate the risk of misjudgment based on a single feature and more reliably determine whether the robot is faulty. The preset alarm threshold is a judgment threshold pre-set based on the industrial robot's safety operation standards, historical fault data, and maintenance requirements. When the fusion support exceeds this threshold, the robot is judged to have malfunctioned.

[0026] In step S202, multi-source data covering the robot's mechanical, electrical, and motion states are collected to ensure data comprehensiveness and support subsequent diagnosis.

[0027] In step S204, targeted processing methods are adopted according to the characteristics of different types of data to eliminate noise interference and extract fault-sensitive features. For example, for vibration signals, wavelet packet decomposition is used to extract the coefficient sequence of specific characteristic frequency bands. For temperature signals, spatial temperature gradient and temporal temperature gradient are calculated. The spatial temperature gradient can identify local overheating of the reducer, and the temporal temperature gradient can capture abnormal temperature rise rate. For current signals, fast Fourier transform is used to extract harmonics from the current signal.

[0028] In step S206, quantitative features are extracted from the augmented data, the abstract signals are transformed into specific indicators that can be used for diagnosis, and integrated into a feature vector. The extracted features include, but are not limited to, vibration features, temperature features, current features, and robot motion features.

[0029] In step S208, based on the DS evidence theory, the degree of support for different fault hypotheses is assigned to each feature subset to achieve feature-fault correlation quantification. First, a fault hypothesis set Θ={H1 (normal), H2 (slight pitting on this wheel), H3 (bearing damage), H4 (motor winding overheating), H5 (motion deviation)} is preset. Then, the feature vector is divided into vibration feature subset, temperature feature subset, current feature subset, and motion feature subset, and the evidence support degree of each subset is calculated.

[0030] In step S210, the multi-source support is integrated through the DS evidence fusion rule to reduce uncertainty and obtain the final diagnostic result.

[0031] Through steps S202 to S210, this application collects multi-source operating status data, and after preprocessing, feature vector construction, multi-feature subset evidence support calculation and fusion, it covers the multi-dimensional operating status of the robot and reduces the risk of misjudgment of a single feature, thereby achieving accurate fault identification and solving the technical problem of low accuracy in robot fault prediction.

[0032] In one optional embodiment, acquiring multiple operational status data of the industrial robot includes at least one of the following: At the first moment, a vibration sampling command is sent to the vibration sensor installed at the input / output flange of the harmonic reducer of the industrial robot to obtain the vibration signal with the sampling timestamp as the target time. At the second moment, a temperature sampling command is sent to the temperature sensor embedded in the motor windings and / or harmonic reducer of the industrial robot to obtain the temperature signal with the sampling timestamp as the target time. At the third moment, a current sampling command is sent to the current transformer of the three-phase input line of the servo drive clamped to the industrial robot to obtain the current signal with the sampling timestamp as the target time. At the fourth moment, a motion sampling command is sent to the motion sensor installed at the joint of the industrial robot to obtain the motion signal with the sampling timestamp as the target time. The first, second, third, and fourth time points are used to ensure that different sensors collect signals according to the same sampling timestamp.

[0033] In this embodiment, the first, second, third, and fourth times are the sampling command issuance times pre-calculated based on sensor data transmission delay and sampling response time. The purpose is to synchronize the actual sampling actions of different sensors and ensure that the sampling timestamps of all data are consistent. The target time is a preset unified sampling timestamp, which serves as the benchmark for determining whether multi-source data are time-synchronized.

[0034] The first moment is for the vibration sensor; the target moment is obtained by subtracting the sum of its sampling response time and transmission delay. The second moment is for the temperature sensor; the target moment is obtained by subtracting the sum of its sampling response time and transmission delay. The third moment is for the current transformer; the target moment is obtained by subtracting the sum of its sampling response time and transmission delay. The fourth moment is for the motion sensor; the target moment is obtained by subtracting the sum of its sampling response time and transmission delay. The motion sensor is a device used to collect motion parameters such as joint output torque and angular velocity, and can monitor abnormal joint loads and motion deviations caused by transmission mechanism jamming.

[0035] In this embodiment, vibration sensors can be installed at the input / output flanges of the harmonic reducer to collect vibration acceleration signals during reducer operation. For example, a MEMS triaxial accelerometer can capture abnormal vibrations from mechanical movements such as gear meshing and bearing rotation. Temperature sensors can be pre-embedded in the motor windings or fixed to the surface of the harmonic reducer to collect temperature data from key components, such as NTC thermistors and thermocouples, to monitor temperature anomalies caused by motor insulation aging and reducer lubrication failure. Current transformers can be clamped onto the three-phase input lines of the servo driver to sense and collect input current signals, capturing current waveform distortions caused by short circuits between motor winding turns or driver failures. Motion sensors can be installed at the joints of an industrial robot to collect motion parameters such as joint output torque and angular velocity.

[0036] This application ensures that the sampling timestamps of multiple data sources, including vibration, temperature, current, and motion, are consistent with the target time by precisely controlling the timing of sampling commands issued by each sensor. This avoids the data misalignment problem caused by time asynchrony in traditional multi-sensor acquisition and improves the accuracy of fault diagnosis.

[0037] In an optional embodiment, before acquiring multiple operational status data of the industrial robot, the method further includes determining the command issuance time of each sensor in the following manner: Step 1: Connect the common test signal to the input of all sampling channels and simultaneously send sampling commands to the sensors of all sampling channels; Step 2: Obtain the capture timestamp of the common test signal captured by the sensor in each sampling channel; Step 3: Calculate the difference between each capture timestamp and the preset reference time, as the sampling compensation for each sensor; Step 4: Obtain the distance from each sensor to the processor, and divide the distance by the signal transmission speed to obtain the distance delay of each sensor; Step 5: Add the preset reference time to the sampling compensation and distance delay of any sensor to obtain the sensor's command issuance time.

[0038] In this embodiment, the common test signal is a standard reference signal with a stable frequency and amplitude, used to uniformly calibrate the delay characteristics of all sampling channels and ensure consistent calibration standards. A sampling channel refers to a complete data acquisition link consisting of a sensor, signal transmission line, and processor input interface; each sensor corresponds to an independent sampling channel. The capture timestamp is the time point at which the sensor completes signal acquisition and recording after receiving the common test signal.

[0039] In this embodiment, the standard common test signal is first connected to the input of all sampling channels through a signal distributor to ensure that the test signal received by each channel is consistent. Then, at a preset reference time, a common test signal sampling command is sent to the sensors of all sampling channels simultaneously, and the preset reference time is recorded.

[0040] In this embodiment, after receiving a sampling command, the sensors in each sampling channel acquire the input common test signal. After acquisition, the acquisition timestamp of the common test signal captured by the sensor is recorded. The acquisition timestamps of all sampling channels are aggregated to the processor to form the raw delay data of each channel. For each sensor, the sampling compensation of that sensor is obtained by subtracting a preset reference time from its acquisition timestamp.

[0041] In this embodiment of the application, for each sensor, the sampling compensation of the sensor is obtained by subtracting the preset reference time from its capture timestamp. The sampling compensation directly reflects the total delay of the sampling channel where the sensor is located from receiving the instruction to completing the acquisition of the common test signal, including the sampling response delay of the sensor itself and the transmission delay of the signal from the sensor to the processor.

[0042] In this embodiment, the physical distance from the installation location of each sensor to the edge processor is measured. Based on the characteristics of the signal transmission medium, such as the electrical signal transmission speed being approximately 77% of the speed of light, the measured distance is divided by the signal transmission speed to obtain the distance delay of each sensor. The distance delay only reflects the pure delay of the signal in the transmission line and is used to correct the line delay error in sampling compensation.

[0043] In this embodiment, the preset reference time is finally added to the sampling compensation and distance delay of the sensor to obtain the command issuance time of the sensor. In this way, the command issuance times of the vibration, temperature, current and motion sensors are calculated respectively to obtain the first time to the fourth time mentioned above.

[0044] The above time synchronization mechanism can be represented as:

[0045] Among them, t sync t represents the time the instruction was issued. base The reference time is the IEEE 1588 protocol reference time, i.e., the preset reference time; d is the distance from the sensor to the processor; c is the signal transmission speed; Δt ADC For sampling compensation.

[0046] It should be noted that the vibration sensor, temperature sensor, current transformer and motion sensor mentioned above can also be installed in other locations on the robot, or other sensors can be used to collect the robot's operating status data.

[0047] This application improves the synchronization accuracy of multi-sensor data from the traditional millisecond level to the microsecond level by using common test signal calibration, sampling compensation and distance delay correction, thus solving the problem of inconsistent delay caused by the differences in characteristics of different sensors.

[0048] In an optional embodiment, the operational status data is preprocessed to obtain enhanced data including at least one of the following: The target wavelet basis function is called to perform wavelet packet decomposition on the vibration signal according to the preset decomposition level, so as to obtain the coefficient sequence of the first characteristic frequency band and the coefficient sequence of the second characteristic frequency band. The first and second characteristic frequency bands are the fault-sensitive frequency bands of the harmonic reducer, and the enhanced data includes the coefficient sequence. Calculate the spatial and temporal temperature gradients of the temperature signal, where the augmented data includes both spatial and temporal temperature gradients; The current harmonics are extracted from the current signal, and the enhanced data includes the current harmonics.

[0049] In this embodiment, the target wavelet basis function is a wavelet basis with good time-frequency localization characteristics, such as the db5 wavelet basis, used for wavelet packet decomposition, which can effectively separate fault-sensitive frequency bands in non-stationary vibration signals. The first characteristic frequency band refers to the 8kHz frequency band, and the second characteristic frequency band refers to the 16kHz frequency band, both of which are fault-sensitive frequency bands of the harmonic reducer. The spatial temperature gradient refers to the ratio of the temperature difference at different temperature measurement points at the same time to the physical distance, reflecting the spatial temperature distribution differences of key components and identifying local overheating. The temporal temperature gradient refers to the ratio of the temperature difference at different times of the same temperature measurement point to the time interval, reflecting the rate of temperature change of key components and identifying abnormal temperature rise. Current harmonics refer to components in the current signal whose frequency is an integer multiple of the fundamental frequency; electrical faults can cause a significant increase in the content of specific harmonics.

[0050] In this embodiment, a vibration signal is acquired, and a target wavelet basis function, such as the db5 wavelet basis, is called. The vibration signal is then decomposed into wavelet packets according to a preset number of decomposition levels, such as 5 levels. During the decomposition process, the original vibration signal is decomposed layer by layer into multiple frequency band sub-bands. The coefficient sequences corresponding to the first and second characteristic frequency bands are extracted from these sub-bands. These coefficient sequences constitute the vibration enhancement data, which can highlight fault characteristics and suppress noise interference. The first characteristic frequency band is, for example, the 8kHz gear meshing frequency band, and the second characteristic frequency band is, for example, the 16kHz bearing fault frequency band.

[0051] Taking the db5 wavelet basis and a decomposition level of 5 as an example, the vibration signal is decomposed into wavelet packets as follows:

[0052] Among them, W j,k [n] refers to the wavelet packet coefficient value at time point n under the j-th level decomposition and k-th frequency band. This refers to the original input signal to be decomposed, where t represents time, and WPD is the wavelet packet decomposition function.

[0053] The decomposition yields the 8kHz frequency band as follows:

[0054] Among them, f s The sampling frequency, such as 16kHz.

[0055] The 16kHz frequency band obtained by decomposition is as follows:

[0056] In this embodiment, temperature signals are acquired. Spatially, the ratio of the temperature difference between adjacent temperature measurement points at the same moment to the physical distance between the two points is calculated to obtain the spatial temperature gradient, which is used to identify localized overheating areas, such as localized high temperatures caused by gearbox lubrication failure. Temporally, the ratio of the temperature difference between the current moment and the previous sampling moment at the same temperature measurement point to the sampling interval is calculated to obtain the temporal temperature gradient, which is used to identify abnormal temperature rise rates, such as accelerated temperature rise caused by motor insulation aging. The spatial temperature gradient and the temporal temperature gradient together constitute temperature enhancement data.

[0057] The space temperature gradient is:

[0058] Among them, T i (t) and T j (t) refers to the temperature values ​​measured at different locations i and j at time t, d ij This refers to the straight-line distance between temperature measuring point i and temperature measuring point j, and the unit is mm.

[0059] The time-temperature gradient is:

[0060] Among them, T bearing (t) refers to the temperature value of the harmonic reducer at the current time t, T bearing (t-Δt) refers to the temperature value of the harmonic reducer at time t-Δt.

[0061] In this embodiment, a current signal is acquired, and a Fast Fourier Transform (FFT) is used to perform frequency domain analysis on the current signal, converting the time-domain current signal into a frequency-domain amplitude spectrum. From the frequency-domain amplitude spectrum, current harmonic components that are integer multiples of the fundamental frequency are extracted, with a focus on harmonics related to electrical faults, such as the 13th harmonic. The extracted current harmonics are the current enhancement data, which can reflect the degree of current waveform distortion.

[0062] Extracting current harmonics:

[0063] Among them, I h (t) is the final output diagnostic characteristic value; the magnitude of this percentage directly reflects the health status of the motor. rms This represents the root mean square value of the current, i.e., the effective value. F{} represents the Fourier transform operator, which transforms a signal from the time domain to the frequency domain. A This represents the time-varying current waveform data directly acquired from the current transformer and serves as the input to the Fourier transform. 13f electThe 13th harmonic represents the fundamental electrical frequency of the motor. The 13th harmonic was chosen because it is particularly sensitive to faults such as broken rotor bars, which can generate unusually strong energy around this frequency.

[0064] This application preprocesses different types of data to highlight fault characteristics, reduce interference, and improve the accuracy of fault diagnosis.

[0065] In an optional embodiment, feature extraction of the augmented data includes at least one of the following: Calculate the energy entropy of the first characteristic frequency band based on the coefficient sequence of the first characteristic frequency band; Calculate the energy entropy of the second characteristic frequency band based on the coefficient sequence of the second characteristic frequency band; The vibration kurtosis of the vibration signal is calculated based on the mean, standard deviation, number of sampling points, and vibration acceleration of the vibration signal. Based on the vibration waveform of the vibration signal, the ratio of the peak value to the effective value of the waveform is calculated as the peak factor. Determine the maximum space temperature gradient based on the space temperature gradient; Calculate the average time-temperature gradient based on the time-temperature gradient; The peak value of the temperature rise acceleration is obtained by calculating the maximum value of the second derivative of the temperature signal with respect to time. Based on current harmonics, calculate the amplitude ratio of the 13th harmonic current; Calculate the total harmonic distortion rate based on current harmonics; Based on current harmonics, determine the content of negative sequence components in the current; Calculate the absolute value of the robot joint output torque in the motion signal; Calculate the absolute value of the robot joint angular velocity in the motion signal; Among them, the energy entropy of the first characteristic frequency band, the energy entropy of the second characteristic frequency band, the vibration kurtosis and the peak factor are the vibration characteristics of the industrial robot; the maximum spatial temperature gradient, the average time temperature gradient and the peak temperature rise acceleration are the temperature characteristics of the industrial robot; the amplitude ratio of the 13th harmonic current, the total harmonic distortion rate and the content of the negative sequence component are the current characteristics of the industrial robot; the absolute value of the robot joint output torque and the absolute value of the robot joint angular velocity are the motion characteristics of the industrial robot.

[0066] In this embodiment, energy entropy refers to the uniformity of signal energy distribution across various frequency bands. Vibration kurtosis describes the steepness of the vibration signal amplitude distribution; fault impacts significantly increase kurtosis, reflecting the intensity of mechanical impact. Peak factor is the ratio of the peak value to the effective value of the vibration signal waveform, used to identify fault impact characteristics in the signal. Maximum spatial temperature gradient is the maximum value in the spatial temperature gradient, reflecting the maximum temperature distribution difference of key components and serving as a direct indicator of local overheating. Average time temperature gradient is the average value of the time temperature gradient, reflecting the average rate of temperature change in key components and being a core indicator of abnormal temperature rise. Peak temperature rise acceleration is the maximum value of the second derivative of the temperature signal with respect to time, reflecting the trend of temperature rise rate changes and predicting the degree of fault deterioration. The 13th harmonic current amplitude ratio is the percentage of the 13th harmonic current amplitude to the fundamental current amplitude; electrical faults significantly increase this ratio. Total harmonic distortion (THD) is the percentage of the root mean square value of all harmonic currents to the root mean square value of the fundamental current, reflecting the degree of current waveform distortion. The negative sequence component content refers to the percentage of the negative sequence component to the positive sequence component of the current. Three-phase imbalance or winding faults in the motor will increase this content. The absolute value of the joint output torque / angular velocity reflects the robot's joint load and motion state; abnormal load or speed indicates a motion control fault.

[0067] In this embodiment of the application, the energy entropy of the first characteristic frequency band (8KHz) can be calculated as follows:

[0068] in, This represents the energy entropy of the 8kHz frequency band, and its magnitude reflects the uncertainty of vibrational energy in the vibratory energy zone around 8kHz. A sudden increase in entropy is a typical characteristic of faults such as gear pitting and tooth surface wear. Ω 8k The set of subbands representing the 8kHz center frequency band is an index set. After 5 levels of wavelet packet decomposition, the signal is divided into 32 equal-width frequency bands. Ω 8k This includes the index numbers of several consecutive sub-bands with a center frequency around 8kHz. W 5,k This refers to the coefficient sequence at the 5th level and kth subband after wavelet packet decomposition. These coefficients represent the strength of the "components" of the original vibration signal within a specific narrow frequency band. k E refers to the energy of a single subband k. It is calculated by first summing the squares of the coefficient sequence, then taking the square root (L2 norm), and finally squaring it again. Essentially, it's about finding the total energy of the coefficient sequence of that subband. k The larger the value, the more intense the signal oscillation in this sub-band. P k This refers to the probability that the energy of the k-th subband accounts for a portion of the total energy. -Σp k lnp k This is the formula for calculating Shannon entropy, when energy is uniformly distributed across all subbands (all p).k When the energy is similar to that of a given subband (a specific p), the entropy is at its maximum; when the energy is concentrated in a certain subband (a specific p), the entropy is at its maximum. k The entropy value is at its minimum when it is close to 1 and the rest are close to 0. In fault diagnosis, an increase in entropy value means that new frequency components have appeared and the energy distribution has become disordered.

[0069] Calculate the energy entropy of the second characteristic frequency band (16kHz). The method is the same as the method for calculating the energy entropy of the first characteristic frequency band (8KHz).

[0070] In this embodiment of the application, the vibration kurtosis of the vibration signal can be calculated as follows:

[0071] Where μ is the mean of the vibration signal, σ is the standard deviation of the vibration signal, representing the fluctuation amplitude of the vibration signal, and N is the number of sampling points of the vibration signal. The vibration acceleration signal is the raw input data, representing a series of acceleration values ​​measured on a certain axis.

[0072] In this embodiment of the application, the peak factor CF of the vibration signal can be calculated by extracting the peak value and effective value of the vibration waveform, calculating the ratio between the two, and identifying the impact characteristics.

[0073] The energy entropy, vibration kurtosis, and peak factor mentioned above together constitute the vibration characteristics of this industrial robot.

[0074] In this embodiment of the application, the temperature characteristics of the industrial robot are as follows:

[0075] in, ΔT refers to the maximum temperature gradient in space, representing the maximum temperature difference between different points on the surface of equipment at the same moment, used to diagnose localized overheating. For example, if the temperature at a certain point on the flange of a harmonic reducer is much higher than the surrounding area, it suggests that the internal gears at that point may have poor lubrication, severe friction, or cracks. This is a direct indicator of localized faults. timeThe average time-temperature gradient represents the rate of temperature change at a point within a unit of time, i.e., the rate of temperature rise, and is used to diagnose persistent overheating faults. If the bearing temperature rises rapidly and continuously within a short period of time, it may indicate a failure of the cooling system, an abnormally increased load, or a failure of the lubricating grease. The average time-temperature gradient reflects the speed of fault development. `max(d²T / dt²)` refers to the maximum value of the second derivative of temperature with respect to time, representing the peak value of the temperature rise acceleration, and is used to diagnose sudden and severe faults. A sudden and drastic acceleration of temperature rise (such as a partial short circuit in a motor winding) will cause this parameter to increase sharply. It is highly sensitive to detecting sudden and severe electrical or mechanical faults (such as short circuits or jamming), enabling very early warning.

[0076] In this embodiment of the application, the current characteristics of the industrial robot are as follows:

[0077] Among them, I 13 The amplitude percentage representing the 13th harmonic current is a gold standard indicator for diagnosing rotor faults. When a motor rotor experiences faults such as broken bars or air gap eccentricity, specific high-order harmonic components (such as the 13th harmonic) will be generated in the current. 13 A significant increase in THD is a clear signal of early subsystem failure. THD, representing Total Harmonic Distortion, measures the degree to which the current waveform deviates from a standard sine wave and can diagnose comprehensive electrical degradation. An increase in THD can be caused by various factors, such as aging of the IGBT switching elements inside the servo driver, performance degradation of the DC bus capacitor, or power grid pollution. neg This represents the content of the negative sequence component in the current. When there is a power supply voltage imbalance, a partial short circuit in the motor windings, or poor phase-to-phase insulation, the negative sequence current will increase significantly. neg It is a key parameter for detecting electrical asymmetry faults.

[0078] In this embodiment of the application, the motion characteristics of the industrial robot are as follows:

[0079] Where |τ| represents the absolute value of the output torque of the robot joint, reflecting the load condition of the joint, |θ| represents the absolute value of the joint angular velocity, and trajectory error represents the deviation between the actual motion trajectory of the robot and the expected trajectory, which is a comprehensive indicator for measuring control accuracy and mechanical accuracy.

[0080] In this embodiment, the vibration characteristics, temperature characteristics, current characteristics, and motion characteristics described above are concatenated to obtain the operational feature vector of the industrial robot:

[0081] This vector represents the operating status information of the industrial robot at a specific moment and will serve as the basis for fault diagnosis.

[0082] This application extracts multiple key features from four dimensions: vibration, temperature, current, and motion. It transforms abstract augmented data into quantitative fault characterization indicators, comprehensively covering possible fault types of industrial robots. This avoids the risk of misjudgment based on a single feature and provides a multi-dimensional and high-quality basis for subsequent evidence support calculation and fusion decision-making, significantly improving the comprehensiveness and accuracy of fault diagnosis.

[0083] In step S208, to calculate the evidence support for different faults for each feature subset in the running feature vector, it is necessary to first set the fault hypothesis set Θ={H1 (normal), H2 (slight pitting on this wheel), H3 (bearing damage), H4 (motor winding overheating), H5 (motion deviation)}, as shown in Table 1: Table 1

[0084] Then, based on the running feature vector F, the basic probability assignment value (BPA) for each feature subset to different hypotheses is calculated, which is the evidence support degree, including vibration evidence BPAs, temperature evidence BPAs, current evidence BPAs, and motion evidence BPAs.

[0085] Vibration evidence BPAs are:

[0086]

[0087] Where, m vib (H2) indicates the degree of support that the current evidence (i.e., vibration characteristics) provides for the H2 hypothesis. vib (H3) represents the degree of support of the current evidence (i.e., vibration characteristics) for the H3 hypothesis. 3.5 is the empirical threshold for vibration kurtosis; when the kurtosis value exceeds 3.5, the system begins to suspect localized damage such as gear pitting. 0.5 is a scaling factor used to adjust the difference between the kurtosis value and the threshold to a suitable range for processing by the sigmoid function. σ(x) is the sigmoid function, used to map a real number to the interval (0,1). When (K... vib When -3.5) is positive and larger, the output is closer to 1, indicating stronger evidence supporting H2. 1.8 is the empirical threshold for 16kHz energy entropy, representing the normal high-frequency energy entropy level. Exceeding this value indicates potential damage such as spalling or cracking in the bearing, resulting in high-frequency impact. 0.8 is the scaling factor.

[0088] Temperature-based evidence BPAs are:

[0089] Where, m temp (H4) represents the degree of support that the current evidence (i.e., temperature characteristics) provides for the H4 hypothesis. 0.2 is the scaling factor. ΔT time The larger the value, the faster the temperature rises and the higher the risk of overheating. time When ΔT is 0, the temperature evidence BPAs are 0, and when ΔT is 0... time When the temperature increases sharply, the temperature evidence BPAs rapidly approach 1, indicating a very high degree of confidence that the winding is overheating.

[0090] The current-based evidence BPAs are:

[0091] Where, m current (H4) represents the degree of support of the current evidence (i.e., current characteristics) for the H4 hypothesis. 0.05 is a preset weighting coefficient used to convert the percentage of harmonic content into its contribution to the BPA. min(0.7,...) is the upper limit cutoff function, indicating that the maximum confidence level for a single current evidence is set at 70%, retaining 30% uncertainty to avoid misjudgment from a single source of evidence, reflecting the advantage of DS theory in handling uncertainty.

[0092] Motion evidence BPAs are:

[0093] Where, m motion (H5) indicates the degree of support of the current evidence (i.e., motion characteristics) for the H5 hypothesis. |τ| is the absolute value of the joint output torque monitored in real time, τ max This is the maximum permissible torque for the joint. This application shifts the center point of the Sigmoid function from 0 to τ. max When the actual torque |τ| is much smaller than τ max When |τ| approaches and exceeds τ, the motion evidence BPAs ≈ 0; max When the BPA value rapidly approaches 1 from 0.5, it indicates that the system is certain that an overload has occurred and there is a risk of motion deviation.

[0094] In step S210, the support levels of all evidence are fused to obtain a fused support level. If the fused support level is greater than a preset alarm threshold, it is determined that the industrial robot has malfunctioned. Specifically, the DS orthogonal sum rule is used to fuse the vibration evidence BPAs, temperature evidence BPAs, current evidence BPAs, and motion evidence BPAs. This combines the trust levels of multiple evidence sources to ultimately generate a fused basic probability allocation m that integrates all evidence sources, has a higher degree of support, and lower uncertainty. fusion (H j ), j=2~5, where the fusion rule is:

[0095]

[0096] When m fusion (H j If the value exceeds a preset alarm threshold, such as 0.5, it is determined that an H-type fault has occurred in the industrial robot. j The corresponding fault.

[0097] In an optional embodiment, after determining that the industrial robot has malfunctioned, the method further includes: At least one of the vibration characteristics, temperature characteristics, current characteristics, and motion characteristics is input into the target physical model to predict the support life of the industrial robot under fault conditions.

[0098] In this embodiment of the application, the target physical model is a remaining life prediction model that integrates physical wear mechanism and data-driven correction. It can output the supportable life (i.e., remaining service life, the remaining time from the current fault state to complete failure, RUL) of key components of industrial robots under fault conditions based on fault characteristics.

[0099] In this embodiment of the application, after determining that the industrial robot has malfunctioned, features strongly related to the malfunction are selected from vibration features, temperature features, current features, and motion features and input into the target physical model. The target physical model outputs the support life and, in conjunction with the production line operation plan, generates targeted maintenance suggestions.

[0100] This application enables lifespan prediction based on fault diagnosis, effectively avoiding sudden downtime losses, reducing resource waste from regular maintenance, and lowering the fault maintenance costs of industrial robots.

[0101] In an optional embodiment, before using the target physical model to predict remaining lifetime, the method further includes obtaining the target physical model in the following manner: Step 1, construct the initial physical model; Step 2: Based on the operating status data, determine at least one of the following as residual inputs: wear increment, temperature gradient, vibration entropy value, and current distortion rate of the industrial robot. Step 3: Input the residual input terms into the Long Short-Term Memory Neural Network Model and obtain the residual correction terms output by the Long Short-Term Memory Neural Network Model; Step 4: Correct the initial physical model using the residual correction term to obtain the target physical model.

[0102] In this embodiment, the initial physical model is a life prediction model built based on the physical wear mechanism of key components of industrial robots (such as harmonic reducers). The residual input terms are parameters reflecting the prediction error of the initial physical model, such as wear increment, temperature gradient, and vibration entropy, used to fine-tune the initial physical model for data-driven correction. The residual correction terms are parameters output by the LSTM residual network used to correct the prediction results of the initial physical model, compensating for complex operating condition interference ignored by the initial model and improving prediction accuracy.

[0103] In the embodiments of this application, such as Figure 3 As shown, residual input terms can be constructed based on data collected by sensors. These residual input terms are then input into an LSTM network to obtain residual correction terms. The residual correction terms are then used to perform data-driven correction on the initial physical model, thereby obtaining the target physical model.

[0104] In this embodiment, the core physical parameters for life prediction can be determined based on the failure physical mechanism of key components of industrial robots. Taking a harmonic reducer as an example, the core physical parameters include the maximum allowable wear of the tooth surface material, the material wear coefficient, the real-time output torque of the joint motor, the joint rotation speed, the friction coefficient, the hardness of the carburized steel tooth surface, and the radius of the reducer flexspline, etc. The initial physical model is constructed as follows:

[0105] Among them, V crit T represents the maximum allowable wear of the tooth surface material of the harmonic reducer, where K is the material wear coefficient, and T is the wear coefficient. in ω is the real-time output torque of the joint motor, μ is the joint rotation speed, μ is the coefficient of friction, H is the hardness of the carburized steel tooth surface, and γ is the radius of the reducer flexure.

[0106] The wear increment, temperature gradient, vibration entropy, and current distortion rate of an industrial robot can be determined based on operational status data and historical operational status data, and residual input terms can be constructed accordingly. X = [Wear increment, Temperature gradient, Vibration entropy, Current distortion rate] Input the residual input terms into the Long Short-Term Memory (LSTM) neural network model, and obtain the residual correction terms output by the LSM model:

[0107] Finally, the initial physical model is corrected using the residual correction term to obtain the target physical model:

[0108] In this embodiment, the initial physical model is corrected using LSTM residual correction terms, which effectively improves the prediction accuracy of the target physical model.

[0109] The complete process of the industrial robot fault diagnosis method provided in this application will be further explained below with reference to section 4.

[0110] like Figure 4 As shown, this application deploys a multimodal sensor network on key components of the industrial robot to collect multi-source heterogeneous data reflecting the operating status of the equipment. For example, a three-axis MEMS vibration sensor is installed at the input and output flanges of the harmonic reducer of the industrial robot to collect vibration signals during the reducer's operation, focusing on capturing vibration characteristics related to mechanical motion such as gear meshing and bearing rotation, providing data support for subsequent judgment of mechanical faults such as reducer wear and bearing cracks. Another example is the pre-embedding of an NTC temperature sensor array inside the robot motor windings, simultaneously covering the temperature measurement points on the harmonic reducer flange, to collect motor winding temperature and reducer flange temperature data, and to monitor the motor insulation status and reducer heat dissipation in real time. Yet another example is clamping a current transformer onto the three-phase input lines of the robot servo driver to collect the driver's input current signal, providing raw data for extracting current harmonics and judging electrical circuit anomalies.

[0111] Multi-source data collected by the multimodal sensor network is transmitted via fieldbus to an edge processor deployed within the robot control cabinet. The edge processor performs several key operations, such as using the db5 wavelet basis function to perform five-level wavelet packet decomposition on the signals collected by the vibration sensor, focusing on 8kHz (sensitive frequency band for gear meshing faults) and 16kHz (sensitive frequency band for bearing defect faults) to enhance the resolution of non-stationary vibration signals and separate fault characteristic frequency components. Another example is the processing of temperature data from motor windings and reducer flanges collected by temperature sensors, calculating spatial and temporal temperature gradients. The spatial temperature gradient reflects the temperature difference between different measurement points, identifying localized abnormal heating, while the temporal temperature gradient reflects the rate of temperature change over time, identifying abnormal temperature rise trends. Finally, harmonic extraction is performed on the current signals collected by the current transformer, calculating the current harmonic content, such as the proportion of the 13th harmonic, to capture the current waveform distortion characteristics caused by electrical system anomalies. Preprocessed data output from the edge processor (such as vibratory subband energy, fault characteristic frequency amplitude, temperature gradient value, and current harmonic content) is transmitted to the feature vector generation module. This module integrates multi-dimensional processing results to construct a multi-dimensional feature vector model. The feature dimensions cover vibration characteristics (such as 8kHz / 16kHz frequency band energy entropy and vibration kurtosis), temperature characteristics (such as maximum temperature and temperature gradient), current characteristics (such as total harmonic distortion (THD) and negative sequence current), and motion characteristics (such as joint torque and motion accuracy deviation), comprehensively characterizing the robot's real-time operating status. The system calculates the evidence support degree for different faults for each feature subset in this feature vector.

[0112] The edge feature vectors are synchronously uploaded to the cloud platform, where a fusion prediction model combining a physical wear model and an LSTM residual network is constructed. The physical wear model calculates the initial remaining life (RULphy) based on physical parameters such as the maximum wear amount of the harmonic reducer material, real-time torque, and friction coefficient. The LSTM residual network learns the residual correction term (ΔLSTM) through historical operating data and finally outputs the corrected final remaining life (RULfinal=RULphy(1+ΔLSTM)). At the same time, the cloud continuously optimizes the model parameters based on the full amount of operating data and generates model update instructions to feed back to the edge processor, ensuring dynamic improvement in diagnostic and prediction accuracy.

[0113] The DS evidence theory fusion engine fuses the support levels of evidence from multiple information sources (or feature groups). The core of the fusion rule is orthogonal summation, which combines the trust levels of multiple evidence sources to ultimately generate a fused basic probability allocation m_fusion(H_j) that integrates all evidence sources, resulting in higher support and lower uncertainty. This result can more comprehensively and reliably reflect the overall operational health status of the equipment, providing a quantitative basis for final fault diagnosis or maintenance decisions (such as early warning, alarm, and recommended shutdown).

[0114] This application provides a method for fault diagnosis of industrial robots, including: acquiring multiple operating status data of the industrial robot; preprocessing the operating status data to obtain enhanced data; extracting features from the enhanced data and constructing an operating feature vector of the industrial robot based on the extracted features; calculating the evidence support degree of each feature subset in the operating feature vector for different faults; fusing all evidence support degrees to obtain fused support degree; and determining that the industrial robot has malfunctioned when the fused support degree is greater than a preset alarm threshold. This application, by collecting multi-source operating status data, preprocessing it, constructing feature vectors, calculating and fusing evidence support degrees of multiple feature subsets, covers multi-dimensional operating states of the robot and reduces the risk of misjudgment based on a single feature, achieving accurate fault identification and solving the technical problem of low accuracy in robot fault prediction.

[0115] According to another aspect of the embodiments of this application, such as Figure 5 As shown, an industrial robot fault diagnosis device is provided, comprising: The acquisition module 501 is used to acquire multiple operating status data of the industrial robot; Preprocessing module 503 is used to preprocess the running status data to obtain enhanced data; Module 505 is used to extract features from the augmented data and construct the operational feature vector of the industrial robot based on the extracted features; Calculation module 507 is used to calculate the evidence support for different faults for each feature subset in the running feature vector; The fusion decision module 509 is used to fuse all evidence support to obtain fusion support, and determine that the industrial robot has malfunctioned if the fusion support is greater than a preset alarm threshold.

[0116] It should be noted that the acquisition module 501 in this embodiment can be used to execute step S202 in this application embodiment, the preprocessing module 503 in this embodiment can be used to execute step S204 in this application embodiment, the construction module 505 in this embodiment can be used to execute step S206 in this application embodiment, the calculation module 507 in this embodiment can be used to execute step S208 in this application embodiment, and the fusion decision module 509 in this embodiment can be used to execute step S210 in this application embodiment.

[0117] It should be noted that the examples and application scenarios implemented by the above modules and corresponding steps are the same, but are not limited to the content disclosed in the above embodiments. It should also be noted that the above modules, as part of the device, can be implemented in software or hardware.

[0118] Optionally, the acquisition module is specifically configured to: send a vibration sampling command to a vibration sensor installed at the input / output flange of the harmonic reducer of the industrial robot at a first moment to obtain a vibration signal with a sampling timestamp as the target moment; send a temperature sampling command to a temperature sensor embedded in the motor windings and / or the harmonic reducer of the industrial robot at a second moment to obtain a temperature signal with a sampling timestamp as the target moment; send a current sampling command to a current transformer clamped to the three-phase input line of the servo driver of the industrial robot at a third moment to obtain a current signal with a sampling timestamp as the target moment; and send a motion sampling command to a motion sensor installed at the joint of the industrial robot at a fourth moment to obtain a motion signal with a sampling timestamp as the target moment; wherein the first, second, third, and fourth moments are used to ensure that different sensors all acquire signals according to the same sampling timestamp.

[0119] Optionally, the acquisition module is further configured to determine the command issuance time of each sensor in the following manner: connect a common test signal to the input of all sampling channels and simultaneously issue sampling commands to the sensors of all sampling channels; acquire the capture timestamp of the common test signal captured by the sensors of each sampling channel; calculate the difference between each capture timestamp and a preset reference time as the sampling compensation for each sensor; acquire the distance from each sensor to the processor and divide the distance by the signal transmission speed to obtain the distance delay of each sensor; add the preset reference time to the sampling compensation and distance delay of any sensor to obtain the command issuance time of the sensor.

[0120] Optionally, the preprocessing module is specifically used for: calling the target wavelet basis function, performing wavelet packet decomposition on the vibration signal according to a preset decomposition level, obtaining the coefficient sequence of the first characteristic frequency band and the coefficient sequence of the second characteristic frequency band, wherein the first characteristic frequency band and the second characteristic frequency band are the fault-sensitive frequency bands of the harmonic reducer, and the enhanced data includes the coefficient sequence; calculating the spatial temperature gradient and temporal temperature gradient of the temperature signal, wherein the enhanced data includes the spatial temperature gradient and the temporal temperature gradient; and extracting current harmonics from the current signal, wherein the enhanced data includes the current harmonics.

[0121] Optionally, this module is specifically used for: calculating the energy entropy of the first characteristic frequency band based on the coefficient sequence of the first characteristic frequency band; calculating the energy entropy of the second characteristic frequency band based on the coefficient sequence of the second characteristic frequency band; calculating the vibration kurtosis of the vibration signal based on the mean, standard deviation, number of sampling points, and vibration acceleration of the vibration signal; calculating the ratio of the peak value to the effective value of the waveform as the peak factor based on the vibration waveform of the vibration signal; determining the maximum spatial temperature gradient based on the spatial temperature gradient; calculating the average temporal temperature gradient based on the temporal temperature gradient; calculating the maximum value of the second derivative of the temperature signal with respect to time to obtain the peak value of the temperature rise acceleration; calculating the amplitude proportion of the 13th harmonic current based on the current harmonics; and calculating the amplitude proportion of the 13th harmonic current based on the current harmonics. The total harmonic distortion rate (THD) is calculated; the content of negative sequence components in the current is determined based on the current harmonics; the absolute value of the robot joint output torque in the motion signal is calculated; the absolute value of the robot joint angular velocity in the motion signal is calculated; among them, the energy entropy of the first characteristic frequency band, the energy entropy of the second characteristic frequency band, the vibration kurtosis, and the peak factor are the vibration characteristics of the industrial robot; the maximum spatial temperature gradient, the average time temperature gradient, and the peak value of the temperature rise acceleration are the temperature characteristics of the industrial robot; the amplitude ratio of the 13th harmonic current, the THD, and the content of negative sequence components are the current characteristics of the industrial robot; the absolute value of the robot joint output torque and the absolute value of the robot joint angular velocity are the motion characteristics of the industrial robot.

[0122] Optionally, the industrial robot fault diagnosis device further includes a life prediction module, specifically used to: input at least one of vibration characteristics, temperature characteristics, current characteristics and motion characteristics into the target physical model, so as to use the target physical model to predict the support life of the industrial robot under fault conditions.

[0123] Optionally, the industrial robot fault diagnosis device further includes a model acquisition module, specifically used for: constructing an initial physical model; determining at least one of the industrial robot's wear increment, temperature gradient, vibration entropy value, and current distortion rate as residual input terms based on operating status data; inputting the residual input terms into a long short-term memory neural network model and obtaining the residual correction terms output by the long short-term memory neural network model; and using the residual correction terms to correct the initial physical model to obtain the target physical model.

[0124] According to another aspect of the embodiments of this application, this application provides an electronic device, such as... Figure 6 As shown, the system includes a memory 601, a processor 603, a communication interface 605, and a communication bus 607. The memory 601 stores a computer program that can run on the processor 603. The memory 601 and the processor 603 communicate through the communication interface 605 and the communication bus 607. When the processor 603 executes the computer program, it implements the steps of the above method.

[0125] The memory and processor in the aforementioned electronic devices communicate with each other via a communication bus and a communication interface. The communication bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into an address bus, a data bus, a control bus, etc.

[0126] The memory may include random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0127] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0128] According to another aspect of the embodiments of this application, a computer program product or computer program is also provided, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the steps of any of the above embodiments.

[0129] Optionally, in embodiments of this application, the computer-readable medium is configured to store program code for the processor to perform the following steps: Acquire multiple operational status data of industrial robots; The operational status data is preprocessed to obtain enhanced data; Feature extraction is performed on the augmented data, and the operational feature vector of the industrial robot is constructed based on the extracted features; Calculate the evidence support for different faults for each feature subset in the running feature vector; All evidence support is fused to obtain fused support, and if the fused support is greater than a preset alarm threshold, it is determined that the industrial robot has malfunctioned.

[0130] Optionally, specific examples in this embodiment can refer to the examples described in the above embodiments, and will not be repeated here.

[0131] In specific implementation, the embodiments of this application can be referred to the above embodiments and have corresponding technical effects.

[0132] It is understood that the embodiments described herein can be implemented in hardware, software, firmware, middleware, microcode, or a combination thereof. For hardware implementation, the processing unit can be implemented in one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), general-purpose processors, controllers, microcontrollers, microprocessors, other electronic units for performing the functions described herein, or combinations thereof.

[0133] For software implementation, the techniques described herein can be implemented by units that perform the functions described herein. The software code can be stored in memory and executed by a processor. The memory can be implemented in the processor or external to the processor.

[0134] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0135] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0136] In the embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative. For instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0137] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0138] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0139] If the aforementioned function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiments of this application, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks. It should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. In the absence of further restrictions, an element defined by the phrase "comprising a..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0140] The above description is merely a specific embodiment of this application, enabling those skilled in the art to understand or implement this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features claimed herein.

Claims

1. A method for diagnosing faults in industrial robots, characterized in that, include: Acquire multiple operational status data of industrial robots; The operational status data is preprocessed to obtain enhanced data; Feature extraction is performed on the enhanced data, and the operational feature vector of the industrial robot is constructed based on the extracted features; Calculate the evidence support for different faults for each feature subset in the running feature vector; All the evidence support scores are fused to obtain a fused support score, and if the fused support score is greater than a preset alarm threshold, it is determined that the industrial robot has malfunctioned.

2. The method according to claim 1, characterized in that, The acquisition of multiple operational status data of the industrial robot includes at least one of the following: At the first moment, a vibration sampling command is sent to the vibration sensor installed at the input / output flange of the harmonic reducer of the industrial robot to obtain the vibration signal with the sampling timestamp as the target time. At the second moment, a temperature sampling command is sent to the temperature sensor embedded in the motor windings and / or the harmonic reducer of the industrial robot to obtain a temperature signal with the sampling timestamp being the target moment. At the third moment, a current sampling command is sent to the current transformer clamped to the three-phase input line of the servo drive of the industrial robot to obtain the current signal with the sampling timestamp of the target moment. At the fourth moment, a motion sampling command is sent to the motion sensor installed at the joint of the industrial robot to obtain a motion signal with the sampling timestamp of the target moment; The first time, the second time, the third time, and the fourth time are used to ensure that different sensors collect signals according to the same sampling timestamp.

3. The method according to claim 2, characterized in that, Before acquiring multiple operational status data of the industrial robot, the method further includes determining the command issuance time of each of the sensors in the following manner: Connect the common test signal to the input of all sampling channels and simultaneously send sampling commands to the sensors of all sampling channels; Obtain the capture timestamp of the common test signal captured by the sensor in each sampling channel; The difference between each of the captured timestamps and the preset reference time is calculated as the sampling compensation for each of the sensors; Obtain the distance from each sensor to the processor, and divide the distance by the signal transmission speed to obtain the distance delay of each sensor; The preset reference time is added to the sampling compensation and distance delay of any of the sensors to obtain the command issuance time of the sensor.

4. The method according to claim 2, characterized in that, The preprocessing of the operational status data to obtain enhanced data includes at least one of the following: The target wavelet basis function is invoked to perform wavelet packet decomposition on the vibration signal according to a preset number of decomposition levels, thereby obtaining the coefficient sequence of the first characteristic frequency band and the coefficient sequence of the second characteristic frequency band, wherein the first characteristic frequency band and the second characteristic frequency band are the fault-sensitive frequency bands of the harmonic reducer, and the enhancement data includes the coefficient sequence; Calculate the spatial temperature gradient and the temporal temperature gradient of the temperature signal, wherein the enhanced data includes the spatial temperature gradient and the temporal temperature gradient; The current harmonics are extracted from the current signal, wherein the enhancement data includes the current harmonics.

5. The method according to claim 4, characterized in that, The feature extraction of the enhanced data includes at least one of the following: Calculate the energy entropy of the first characteristic frequency band based on the coefficient sequence of the first characteristic frequency band; Calculate the energy entropy of the second characteristic frequency band based on the coefficient sequence of the second characteristic frequency band; The vibration kurtosis of the vibration signal is calculated based on the mean, standard deviation, number of sampling points, and vibration acceleration of the vibration signal. Based on the vibration waveform of the vibration signal, the ratio of the peak value to the effective value of the waveform is calculated as the peak factor; The maximum space temperature gradient is determined based on the aforementioned space temperature gradient. Calculate the average time-temperature gradient based on the aforementioned time-temperature gradient; The peak value of the second derivative of the temperature signal with respect to time is obtained by calculating the maximum value of the temperature rise acceleration. Based on the current harmonics, calculate the amplitude ratio of the 13th harmonic current; Calculate the total harmonic distortion rate based on the current harmonics; Based on the current harmonics, the content of the negative sequence component in the current is determined; Calculate the absolute value of the robot joint output torque in the motion signal; Calculate the absolute value of the robot joint angular velocity in the motion signal; Wherein, the energy entropy of the first characteristic frequency band, the energy entropy of the second characteristic frequency band, the vibration kurtosis, and the peak factor are the vibration characteristics of the industrial robot; the maximum spatial temperature gradient, the average time temperature gradient, and the peak temperature rise acceleration are the temperature characteristics of the industrial robot; the amplitude ratio of the 13th harmonic current, the total harmonic distortion rate, and the content of the negative sequence component are the current characteristics of the industrial robot; the absolute value of the robot joint output torque and the absolute value of the robot joint angular velocity are the motion characteristics of the industrial robot.

6. The method according to claim 5, characterized in that, After determining that the industrial robot has malfunctioned, the method further includes: The vibration characteristics, temperature characteristics, current characteristics, and motion characteristics are input into the target physical model to predict the sustainable life of the industrial robot under fault conditions.

7. The method according to claim 6, characterized in that, Before using the target physical model to predict remaining lifetime, the method further includes obtaining the target physical model in the following manner: Construct an initial physical model; Based on the operating status data, at least one of the following is determined as the residual input: wear increment, temperature gradient, vibration entropy value, and current distortion rate of the industrial robot. The residual input term is input into the long short-term memory neural network model, and the residual correction term output by the long short-term memory neural network model is obtained; The initial physical model is corrected using the residual correction term to obtain the target physical model.

8. An industrial robot fault diagnosis device, characterized in that, include: The acquisition module is used to acquire multiple operating status data of the industrial robot; The preprocessing module is used to preprocess the running status data to obtain enhanced data; A construction module is used to extract features from the enhanced data and construct the operational feature vector of the industrial robot based on the extracted features; The calculation module is used to calculate the evidence support degree of each feature subset in the running feature vector for different faults; The fusion decision module is used to fuse all the evidence support scores to obtain a fusion support score, and determine that the industrial robot has malfunctioned if the fusion support score is greater than a preset alarm threshold.

9. An electronic device comprising a memory, a processor, a communication interface, and a communication bus, wherein the memory stores a computer program executable on the processor, and the memory and the processor communicate via the communication bus and the communication interface, characterized in that, When the processor executes the computer program, it implements the industrial robot fault diagnosis method according to any one of claims 1 to 7.

10. A computer-readable medium having processor-executable non-volatile program code, characterized in that, The program code causes the processor to execute the industrial robot fault diagnosis method according to any one of claims 1 to 7.

Images