A diagnostic method for robot arm failure detection

Abstract

The application discloses a kind of machine arm fault detection diagnostic methods, it is related to machine arm field, comprising the following steps: collecting the vibration signal of machine arm;Washing collected vibration signal data;Utilize empirical mode decomposition algorithm to process washed data, obtain N intrinsic mode function, and extract time domain feature and frequency domain feature, carry out normalization processing;Training sample set, field working condition sample set are constructed;Abnormal detection model based on deep convolutional neural network is constructed;Field working condition sample set is input into abnormal detection model, and health state value is returned;Health state value is returned to the machine arm fault diagnosis system software developed, and the health state real-time detection of machine arm is realized.The application utilizes abnormal detection model based on deep convolutional neural network, realizes the development of machine arm fault diagnosis system, improves the efficiency of machine arm fault detection, ensures the normal operation of production line by timely feedback machine arm operating state.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of robot arm, and particularly relates to a diagnosis method for robot arm fault detection. BACKGROUND

[0002] The robot arm is widely used in the fields of semiconductor manufacturing, biopharmaceutical, industrial metallurgy, space exploration, etc. It is a mechanical device capable of automatically completing the operation of objects (such as workpieces, parts, materials, etc.) according to the established program or requirements. It has been used in many countries for pipeline, workshop and other places to replace manual work to complete welding, painting, handling, cutting and other high-strength and high-risk tasks. As an important part of industrial production, the robot arm will have a huge impact on the entire industrial production or pipeline and cause incalculable losses in all aspects when any fault occurs. Therefore, it is a valuable research topic to design an effective fault detection system to provide timely feedback when the robot arm system has a problem, avoid further deterioration, and ensure the reliable stability of the robot arm system during operation.

[0003] At present, the main diagnosis method for the robot arm system still remains in the stage of detecting the physical information of each joint, which mainly includes two methods: one is to measure the difference between the rotation angle of the robot arm joint and the preset value to determine whether a fault occurs; the other is to establish a substitute connecting rod of the robot arm joint, and to compare the length residual error between the virtual connecting rod calculated and the actual connecting rod to determine whether a fault occurs. Both of the two methods are model-based fault detection methods. However, the robot arm system is actually a very complex nonlinear system, so the fault diagnosis method based on the simplified model has not achieved particularly ideal results in practical application. In order to overcome this problem, in recent years, the robot arm fault diagnosis method based on the model method as the main body and other diagnosis methods as the auxiliary method has also been continuously proposed. However, since these methods still use the residual error generation and estimation method for fault diagnosis, they still belong to the model-based fault diagnosis category, and the actual application is still relatively small.

[0004] Based on the technical features of fault detection, the fault diagnosis method can be roughly divided into the following three types: fault diagnosis method based on analytical model, fault diagnosis method based on signal and fault diagnosis method based on knowledge. At present, the main diagnosis method of the robot arm system belongs to the first type, that is, the fault diagnosis method based on analytical model. However, with the vigorous development of artificial intelligence, more and more machine learning algorithms are introduced into industrial production and manufacturing, and the third type of fault diagnosis method based on knowledge is increasingly valued. In the field of fault diagnosis of nonlinear systems, this method has unique advantages and feasibility. Therefore, introducing the anomaly detection algorithm in machine learning, such as support vector machine (SVM), isolated forest (IF) and auto-encoding neural network (AE), into the fault detection of robot arm system can effectively promote the development of the field. However, in the actual problem of robot arm system fault detection, the health data is much more than the abnormal data, and even in some specific scenarios, there is a lack of abnormal data. In addition, due to the difference of robot arms in different scenes, the consistency of data between different devices is poor, and it is difficult to realize model transfer learning. Many existing machine learning anomaly detection algorithms are difficult to solve the above two key problems.

[0005] Therefore, the skilled in the art is committed to developing a robot arm fault detection diagnosis method, which can realize the development of robot arm fault diagnosis system based on machine learning algorithm, can be applied to different robot arm systems, and can more comprehensively and effectively detect robot arm system. SUMMARY

[0006] In view of the above defects of the prior art, the technical problem to be solved by the present application is how to develop a robot arm fault detection model based on a deep neural network and establish a complete fault diagnosis system.

[0007] To achieve the above object, the present application provides a robot arm fault detection diagnosis method, which comprises the following steps:

[0008] Step 1, vibration sensors are arranged at multiple key positions of the robot arm, and vibration signals of the robot arm are collected;

[0009] Step 2, the vibration signal data is preprocessed;

[0010] The step 2 further comprises:

[0011] Step 2.1, the collected vibration signal data is cleaned, and working state data is screened out;

[0012] Step 2.2, the cleaned data is processed by using empirical mode decomposition (EMD) algorithm to obtain N intrinsic mode functions (IMF); for each intrinsic mode function (IMF), time domain features and frequency domain features are extracted and normalized;

[0013] Step 3, using the pre-processed health data to construct a training sample set, using the pre-processed field working condition data to construct a field working condition sample set;

[0014] Step 4, constructing an anomaly detection model based on a deep convolutional neural network;

[0015] Step 5, inputting the field working condition sample set into the trained deep convolutional neural network anomaly detection model, and returning a health state value;

[0016] Step 6, returning the health state value to the developed robot arm fault diagnosis system software to realize real-time detection of the health state of the robot arm.

[0017] Further, the key positions include robot arm joints, servo motors and reducers.

[0018] Further, the step 2.1 further comprises: judging the running state of the robot arm system by extracting the maximum amplitude in the time domain of the vibration signal data, and screening out the vibration signal data in the working state; the running state includes working state, rest state.

[0019] Further, the time domain features in the step 2.2 include: maximum value Max, mean value Mean, peak-to-peak value PP, root mean square value RMS, standard deviation Std, variance Var, skewness SK, kurtosis KU, waveform indication SH, pulse index IM, peak value index CR, and margin index CL. The 12 time domain features construct a time domain feature vector:

[0020] X time =[Max,Mean,PP,RMS,Std,Var,SK KU,SH,IM,CR,CL]

[0021] Further, the frequency domain features in the step 2.2 include: spectral energy FE, center of gravity frequency FC, mean square frequency MSF, root mean square frequency RMSF, frequency variance VF, and frequency standard deviation RVF. The 6 frequency domain features construct a frequency domain feature vector:

[0022] X fre =[FE,FC,MSF,RMSF,VF,RVF]

[0023] Further, the time domain features and the frequency domain features are combined to obtain an 18-dimensional feature vector X of the vibration signal X time ,Xfre The normalization processing refers to mapping each feature dimension of the 18-dimensional feature set vector X = [X time , X fre ] into [0, 1].

[0024] Further, the step 4 further comprises:

[0025] Step 4.1, dividing the training sample set composed of health data into a training data set {X i} train and a verification data set {X i} validation according to the proportions of 80% and 20%, and giving the training data set {X i} train a real label "0";

[0026] Step 4.2, building a deep convolutional neural network CNN, including three convolutional layers and three fully connected layers; using a mean square error function to construct a loss function between the output of the deep convolutional neural network and the real label of the data set;

[0027] Step 4.3, using a Gaussian distribution N(0, var) to initialize the parameters of the deep convolutional neural network; using a stochastic gradient descent algorithm to optimize the loss function to train the model until the loss function is reduced to 1x10 -4 ;

[0028] Step 4.4, inputting the verification data set {X i} validation into the deep convolutional neural network to obtain the output {Y i} validation of the verification data set, and taking the absolute value to obtain {|Y| i} validation ;

[0029] Step 4.5, judging whether the absolute value {|Y| i} validation is less than a given threshold value; if the absolute value {|Y| i} validation is less than the given threshold value, it is considered that the trained deep convolutional neural network-based anomaly detection model meets the expected effect, the model and its parameters are saved, and the process is ended; if the absolute value {|Y| i} validation is greater than the given threshold value, it is considered that the expected effect has not been reached, the model and the initialization parameters are changed, and the step 4.3 is repeated.

[0030] Further, the step 4.3 further comprises: var in the Gaussian distribution N(0, var) needs to take a very large number, which aims to reduce the generalization performance of the neural network by large initialization parameters.

[0031] Further, the step 5 further comprises:

[0032] Step 5.1, the field working condition sample set {X i} test is input into the saved deep convolutional neural network, and the field working condition sample set output {Y i} test is obtained, and further, the absolute value thereof is obtained {|Y| i} test .

[0033] Step 5.2, the health state value of the field working condition sample set is calculated, and is defined as follows:

[0034]

[0035] Wherein, represents the average error of the field working condition sample set, represents the average error of the verification set.

[0036] Further, the step 6 further comprises: the machine arm fault diagnosis system software receives the calculated health state value of the field working condition sample set in real time, and displays it on the software system page; if the health state value is too large, the machine arm fault diagnosis system software will timely issue an alarm to remind the field engineer of the possible fault; the software can also backtrack to find the sensor where the vibration data causing the health state value to be too large, and remind the field engineer of the specific part of the machine arm that may fail and the type of failure.

[0037] Compared with the prior art, the present application has at least the following beneficial technical effects:

[0038] 1、The present application uses a newly developed abnormality detection model based on a deep convolutional neural network to realize the development of a machine arm fault diagnosis system, which improves the efficiency of machine arm fault detection on the one hand, and can timely feedback the running state of the machine arm to ensure the normal operation of the production line on the other hand;

[0039] 2、The present application installs vibration sensors at multiple vulnerable parts of the machine arm system for comprehensive fault detection, which can improve the detection efficiency and accuracy;

[0040] 3、The application judges the operation state of the machine arm system by extracting features from the collected data, improves the efficiency of sensor data collection, improves the working efficiency of the machine arm fault diagnosis system, saves computing resources, and even reduces the risk of misjudgment;

[0041] 4、The application improves the effectiveness and reliability of feature extraction by stabilizing the vibration signal, thereby improving the stability and effectiveness of the model;Through selecting appropriate feature quantities, the working effectiveness of the machine arm system fault diagnosis system and the abnormality detection model based on deep convolutional neural network is improved, and the fault of the machine arm can be captured by the algorithm model through the signal collected by the vibration sensor;

[0042] 5、The abnormality detection model based on deep convolutional neural network constructed by the application will not encounter the problem of dimension disaster, can process higher-dimensional data problems, support multi-sensor collaborative modeling analysis, and fault detection;The abnormality detection model based on deep convolutional neural network belongs to a knowledge-based fault detection model, which often has better performance when processing complex nonlinear systems;The abnormality detection model based on deep convolutional neural network fully utilizes the characteristics of deep neural networks with large initialization, which are poor in generalization and not easy to learn zero mapping, so that the neural network can directly process single classification problems, and then an abnormality detection model is constructed;The abnormality detection model based on deep convolutional neural network provides a way for neural networks to do single classification problems and abnormality detection in addition to Auto-encoder.

[0043] The concept, specific structure and technical effects of the application will be further described below with reference to the accompanying drawings, so as to fully understand the purpose, features and effects of the application. BRIEF DESCRIPTION OF DRAWINGS

[0044] Figure 1 is a method flowchart of a preferred embodiment of the application;

[0045] Figure 2 is an algorithm construction and training schematic diagram of an abnormality detection model based on deep convolutional neural network of a preferred embodiment of the application;

[0046] Figure 3 is a machine arm health value change graph in a factory of a preferred embodiment of the application. DETAILED DESCRIPTION

[0047] The application can be embodied in many different forms, and the scope of protection of the application is not limited to the embodiments described herein.

[0048] In the drawings, components of the same structure are designated with the same reference numerals, and components similar in structure or function are designated with like reference numerals. The size and thickness of each component shown in the drawings are arbitrarily shown, and the size and thickness of each component are not limited in the present application. In order to make the drawings clearer, the thickness of some components is exaggerated in some places in the drawings.

[0049] As Figure 1 shown is a method flowchart of a preferred embodiment of the present application, including the following steps:

[0050] S1, vibration sensors are arranged at multiple key positions of the robot arm, such as joints of the robot arm, servo motors, and reducers, and vibration signals of the robot arm in normal working state are collected.

[0051] The selection of sensor positions is very important. The conventional robot arm fault diagnosis system mainly uses information at the joints of the robot arm, and we install vibration sensors at key positions of the robot arm, such as joints, servo motors, and reducers, and receive signals, which not only realizes full-fledged fault detection of the robot arm system, but also grasps details that are easily overlooked in the conventional method. Using the all-around information provided by these sensors helps the fault diagnosis system to accurately feedback the running state of the robot arm.

[0052] In S1, vibration signals of the robot arm in normal working state are also collected.

[0053] S2, the collected vibration signal data is cleaned, and working state data is selected.

[0054] The data cleaning mainly uses the feature that the amplitude of the vibration signal data of the robot arm system in the working state and the resting state is obviously different, and judges the running state by extracting the amplitude maximum Max in the time domain of the signal data, that is, if the value is large, it is considered to be in the working state; otherwise, if it is small, it is considered to be in the resting state. When collecting data, by extracting the characteristics of the data, the running state of the robot arm system is judged, the vibration signal data in the working state is selected, the efficiency of data collection is improved, the consumption of computing resources is reduced, and even the risk of misjudgment is reduced

[0055] S3, the cleaned data is processed by using the empirical mode decomposition (EMD) algorithm to obtain N intrinsic mode functions (IMF). Then, for each intrinsic mode function (IMF), time domain features and frequency domain features are extracted and normalized.

[0056] We use empirical mode decomposition (EMD) algorithm to process the cleaned data in S2, get N intrinsic mode function (IMF) data, and realize the stationarization of vibration signal data. Empirical mode decomposition has obvious advantages in processing non-stationary and nonlinear data, is suitable for analyzing nonlinear and non-stationary signal sequences, and has very high signal-to-noise ratio. Therefore, using empirical mode decomposition (EMD) algorithm to process the non-stationary signal of the robot arm system, extracting the stationary signal data, and then performing feature extraction, can obtain more effective and reliable results.

[0057] After obtaining N intrinsic mode function (IMF) data, time domain features and frequency domain features are extracted for each intrinsic mode function. The time domain features include: maximum value Max, mean value Mean, peak-to-peak value PP, root mean square value RMS, standard deviation Std, variance Var, skewness SK, kurtosis KU, waveform indicator SH, pulse index IM, peak value index CR, and margin index CL. These 12 common time domain features can also be selected, and the 12 time domain features are constructed into a time domain feature vector X time =[Max,Mean,PP,RMS,Std,Var,SK KU,SH,IM,CR,CL]; The frequency domain features include: spectral energy FE, center of gravity frequency FC, mean square frequency MSF, root mean square frequency RMSF, frequency variance VF, and frequency standard deviation RVF. These 6 common frequency domain features are constructed into a frequency domain feature vector X fre =[FE,FC,MSF,RMSF,VF,RVF]. Finally, the time domain features and frequency domain features are combined to obtain an 18-dimensional feature vector X of the vibration signal X time =[X fre ]. In the 18-dimensional feature vector, the most critical are the maximum value Max of the time domain feature and the center of gravity frequency FC of the frequency domain feature. The former can effectively reflect the change of the amplitude of the vibration signal with time, and the latter can effectively reflect the change of the frequency of the vibration signal with time.

[0058] Normalization processing refers to mapping each feature dimension of the 18-dimensional feature set vector X time =[X fre ] to [0, 1]. Normalization processing can improve the calculation efficiency of the algorithm and improve the judgment effect of the system.

[0059] S4, using the pre-processed (i.e., after S2, S3) health data to construct a training sample set, and using the pre-processed field working condition data to construct a field working condition sample set. Further, the training sample set is divided into a training set and a validation set.

[0060] S5, build the abnormality detection model based on deep convolutional neural network, train the model through the training set and stochastic gradient descent (SGD) algorithm, and ensure that the model reaches the final effect through the validation set. Finally, we get the trained abnormality detection model based on deep convolutional neural network, which is ready for further construction of fault diagnosis model. As shown in FIG. 8, it is an algorithm construction and training diagram of the abnormality detection model based on deep convolutional neural network, including the following steps: Figure 2

[0061] S5.1, the training set {X i} healthy composed of healthy data is divided into training data set {X i} train and validation data set {X i} validation according to the proportion of 80% and 20%, and the training data set {X i} train is given a real label "0".

[0062] S5.2, build a deep convolutional neural network CNN, which is composed of three convolutional layers and three fully connected layers, wherein the convolutional layer uses the same size padding, and each hidden layer of the fully connected layer is provided with 1024 neurons. Assuming that in S3, after the empirical mode decomposition (EMD) algorithm is processed, N intrinsic mode function (IMF) data are obtained, and K features are extracted from each intrinsic mode function data, then the input size of the deep convolutional neural network is a two-dimensional tensor of N×K. The output of the deep convolutional neural network is a one-dimensional scalar. The mean square error function is used to construct the loss function between the output of the deep convolutional neural network and the real label of the data set.

[0063] S5.3, the parameters of the deep convolutional neural network are initialized using Gaussian distribution N(0,var). Here, var needs to take a very large number to reduce the generalization performance of the deep convolutional neural network. Then, the random gradient descent algorithm is used to optimize the loss function to train the model until the loss function is reduced to 1×10 -4 .

[0064] S5.4, input the validation data set {X i} validation into the deep convolutional neural network to obtain the output {Y i} validation of the validation data set, and further take the absolute value to obtain {|Y| i} validation ​. Determine whether the absolute value is less than a given threshold value TV1: if less than the given threshold value, consider that the trained deep convolutional neural network-based anomaly detection model meets the expected effect, save the model and its parameters, and end; if greater than the threshold value, consider that the expected effect has not been reached, change the model and the initialization parameters, and repeat step S5.3.

[0065] S6, input the field working condition sample set into the trained deep convolutional neural network-based anomaly detection model, return a health state value (Healthy Index, HI), and realize fault detection and health management of the robot arm system.

[0066] S6.1, input the field working condition sample set {X i} test into the saved deep convolutional neural network, obtain the field working condition sample set output {Y i} test , and further take the absolute value to obtain { |Y| i} test .

[0067] S6.2, calculate the health state value (Healthy Index, HI) I healthy of the field working condition sample set, which is defined as follows:

[0068]

[0069] wherein, represents the average error of the field working condition sample set, represents the average error of the verification set. Generally, when I healthy is greater than or equal to 1, it is considered that the field working condition data set has obvious faults. I healthy In essence, it measures the size of the average error of the field working condition sample set compared with the average error of the verification set. If the average error of the working condition sample set is much larger than the average error of the verification set, it is considered that the field working condition data set has faults; otherwise, if the two sizes are similar, it is considered that no faults occur.

[0070] S7, return the health state value to the developed robot arm fault diagnosis system software, realize real-time state detection, and remind the field engineer of the possible fault position and type when a possible fault occurs.

[0071] The robot arm fault diagnosis system software will receive the calculated health state value (Healthy Index, HI) I healthy of the field working condition sample set in real time, and display it on the software system page. If the health state value I healthyIf the vibration data is too large, the robot arm fault diagnosis system software will timely issue an alarm to remind the on-site engineer of possible faults. At the same time, the software will also backtrack to find the sensor where the vibration data is too large, remind the on-site engineer of the specific part of the robot arm that may have a fault and the type of fault. In this way, intelligent detection and health management of the robot arm system are achieved. healthy healthy healthy healthy

[0072] In order to verify the feasibility of the robot arm fault detection system proposed in the present application, we use the data collected by the robot arm operating in the factory to conduct experimental verification. The sampling frequency of the vibration sensor is 4000 Hz, and the sampling is performed for 13 days. We use the vibration data of the first two days as the health data set, and the data set of the last 11 days as the on-site engineering data set, and perform model construction, training and analysis, and return the corresponding health state value. During this period, the on-site professional engineer artificially created a small fault on the 9th day and reported the occurrence of the fault, and finally obtained the result in Table 1. Figure 3

[0073] In Table 1, the abscissa is the day, and the ordinate is the health state value I healthy of the corresponding day. healthy We found that first, during the 1st-8th day, the health state value I healthy is very low, remaining below 0.1, which indicates that the robot arm is in a normal operating state and is healthy during these 8 days; and on the 9th day, the health state value I suddenly soared to above 0.7, which indicates that the robot arm had a fault on the 9th day and was abnormal. This result is very consistent with the actual on-site situation, i.e., the on-site engineer reported normal operation during the 1st-8th day, and reported an abnormality by artificially creating a small fault on the 9th day.

[0074] In summary, the robot arm system fault detection system based on the convolutional neural network provided in the present application is innovative and practical, and can solve the problem of robot arm fault detection.

[0075] The above describes in detail the preferred embodiments of the present application. It should be understood that those skilled in the art can make many modifications and changes to the present application without creative labor based on the concept of the present application. Therefore, any technical solution obtained by logical analysis, reasoning or limited experiment based on the prior art according to the concept of the present application shall be within the scope of protection determined by the claims.

Claims

1. A diagnostic method of machine arm failure detection, characterized by, The method comprises the following steps: Step 1, setting vibration sensors at multiple key positions of the robot arm and collecting vibration signals of the robot arm; Step 2, preprocessing the vibration signal data; The step 2 further comprises: Step 2.1, cleaning the collected vibration signal data to screen out working state data; Step 2.2, processing the cleaned data by using an empirical mode decomposition algorithm to obtain N intrinsic mode functions; for each intrinsic mode function, time domain features and frequency domain features are extracted and normalized; Step 3, constructing a training sample set by using the preprocessed health data and constructing a field working condition sample set by using the preprocessed field working condition data; Step 4, constructing an abnormality detection model based on a deep convolutional neural network; Step 5, inputting the field working condition sample set into the trained deep convolutional neural network abnormality detection model to return a health state value; Step 6, returning the health state value to the developed robot arm fault diagnosis system software to realize real-time detection of the health state of the robot arm; The step 4 further comprises: Step 4.1, Divide the training sample set constituted by the health data into a training data set and a verification data set according to a ratio of 80% and 20% and a verification data set , and give the training data set a true label "0"; Step 4.2, building a deep convolutional neural network CNN, including three convolutional layers and three fully connected layers; using a mean square error function to construct a loss function between the output of the deep convolutional neural network and the true label of the data set; Step 4.3, initialize the parameters of the deep convolutional neural network using a Gaussian distribution N(0, var); train the model using a stochastic gradient descent algorithm to optimize the loss function until the loss function is reduced to 1 x 10 -4 . Step 4.4: Validate the dataset The data is fed into a deep convolutional neural network to obtain the output of the validation dataset. and take its absolute value. ; Step 4.5, judging the absolute value whether it is less than a given threshold value; if the absolute value is less than the given threshold value, it is considered that the trained deep convolutional neural network-based anomaly detection model meets the expected effect, the model and its parameters are saved, and the process ends; if the absolute value is greater than the given threshold value, it is considered that the expected effect has not been reached, the model and the initial parameters are changed, and the step 4.3 is repeated.

2. The diagnostic method of machine arm fault detection according to claim 1, wherein, The key positions in the step 1 include robot arm joints, servo motors and reducers.

3. The diagnostic method of machine arm fault detection according to claim 1, wherein, The step 2.1 further comprises: judging the running state of the robot arm system by extracting the maximum amplitude in the time domain of the vibration signal data to screen out the vibration signal data in the working state; the running state includes the working state and the resting state.

4. The diagnostic method of machine arm fault detection according to claim 1, wherein, The time domain features in the step 2.2 include: maximum value Max, mean value Mean, peak-to-peak value PP, root mean square value RMS, standard deviation Std, variance Var, skewness SK, kurtosis KU, waveform indicator SH, pulse indicator IM, peak value indicator CR and margin indicator CL; the 12 time domain features construct a time domain feature vector: 。 5. The diagnostic method of machine arm fault detection according to claim 4, wherein, The frequency domain features in the step 2.2 include: spectral energy and FE, center of gravity frequency FC, mean square frequency MSF, root mean square frequency RMSF, frequency variance VF and frequency standard deviation RVF; the 6 frequency domain features construct a frequency domain feature vector: 。 6. The diagnostic method of machine arm fault detection according to claim 5, wherein, The time domain features are combined with the frequency domain features to obtain an 18-dimensional feature vector of the vibration signal ; the normalization processing refers to mapping each feature dimension of the 18-dimensional feature set vector to [0, 1].

7. The diagnostic method of machine arm fault detection according to claim 6, wherein, The step 4.3 further comprises: var in the Gaussian distribution N(0, var) needs to take a large number, and the purpose is to reduce the generalization performance of the neural network by large initialization parameters.

8. The diagnostic method of machine arm fault detection according to claim 1, wherein, The step 5 further comprises: Step 5.1, inputting the set of field condition samples into a saved deep convolutional neural network to obtain a set of field condition sample outputs and further taking the absolute value thereof to obtain ;​ Step 5.2, calculating the health state value of the field working condition sample set, which is defined as follows: wherein, represents the average error of the field condition sample set, represents the average error of the validation set.

9. The diagnostic method of robot arm fault detection according to claim 1, characterized in that, The step 6 further comprises: the robot arm fault diagnosis system software receives the calculated health state value of the field working condition sample set in real time and displays it on the software system page; if the health state value is too large, the robot arm fault diagnosis system software will timely alarm to remind the field engineer of the possible fault; the software can also backtrack to find the sensor where the vibration data causing the large health state value is located, and remind the field engineer of the specific position and type of the possible robot arm fault.

Images