KR102449413B1 - Apparatus and method for detecting fault of robot - Google Patents

Abstract

The present invention relates to an apparatus and method for detecting a failure of a robot, and the apparatus for detecting a failure of a robot according to an embodiment includes a current corresponding to at least one of a mechanical defect, an electronic defect, and an electrical defect from at least one motor installed in the robot. A data collection unit collecting data, a data preprocessing unit performing preprocessing for applying a dimension reduction technique to the collected current data, and extracting feature data according to at least one defect based on wavelet analysis on the preprocessed data It includes a feature extraction unit and a failure detection unit for detecting a failure of the robot based on the extracted characteristic data.

Description

Fault detection device and method of robot

The present invention relates to an apparatus and method for detecting a failure of a robot, and more particularly, to a technical idea for detecting and diagnosing a failure of a robot for PHM of an industrial robot.

Modern industrial systems have minimized human interference by applying robots programmed to perform automated tasks along with advances in automation and control. Because the robot, which is the most basic component of the automation of industrial systems, is continuously used in the manufacturing process, the performance of the robot's partial system or parts deteriorates over time.

Failure to do proper maintenance when such degradation occurs can lead to various failures in the system, which in turn can lead to unexpected plant downtime and loss of production.

To solve this, prognostics and health management (PHM), which performs condition monitoring, diagnosis, prediction, and maintenance schedules, is required.

The most important technology of PHM is fault detection and diagnosis. A method of detecting and diagnosing a failure of an electrical device uses a current signal, and detecting a mechanical defect through a vibration or acoustic emission signal is used.

This makes the overall fault detection and diagnosis method more complex, and the handling of multiple types of data (current, vibration, lubricating oil analysis or acoustic emissions) makes it difficult to manage in real-time implementations and increases the computational cost of the entire system.

In addition, the accuracy of fault diagnosis using vibration analysis tends to be low at low speed of the motor and at the initial stage of start-up. Therefore, it is necessary to establish a systematic approach to discriminate various types of defects using only one data for electrical and mechanical defects.

Korean Patent Laid-Open Patent No. 10-2005-0066765, "Online Diagnosis Method of Motor Using MCSA" Korean Patent Laid-Open Patent No. 10-2010-0112734, "On-site complex fault diagnosis method of induction motor"

An object of the present invention is to provide a robot failure detection apparatus and method that can easily detect and diagnose mechanical failure, electronic failure, and electrical failure of the robot through motor current signature analysis (MCSA) and machine learning. .

Another object of the present invention is to provide an apparatus and method for detecting a failure of a robot that can accurately classify defects according to mechanical failure, electronic failure, and electrical failure of the robot using only the three-phase current of the electric motor.

Another object of the present invention is to provide a failure detection apparatus and method capable of reducing a calculation time in real-time implementation to have a fast response speed, and greatly reducing sensor components, thereby reducing failure detection and diagnosis costs.

A robot failure detection apparatus according to an embodiment of the present invention includes a data collection unit for collecting current data corresponding to at least one of a mechanical defect, an electronic defect, and an electrical defect from at least one motor installed in the robot; A data preprocessor that performs preprocessing for applying a dimensionality reduction technique to current data, a feature extractor that extracts feature data according to at least one defect based on wavelet analysis on the preprocessed data, and a feature extractor based on the extracted feature data It may include a failure detection unit for detecting a failure of the robot.

According to one side, the data collector may collect three-phase current data of at least one motor derived through motor current signature analysis (MCSA).

According to one side, the data preprocessor may dimensionally reduce the three-phase current data to the two-phase current data by using DQ0 transformation.

According to one side, the feature extractor may decompose preprocessed data into wavelet signals through time-domain analysis based on discrete wavelet transform (DWT).

According to one side, the feature extractor may extract the first feature by performing at least one of wavelet domain analysis and wavelet statistical analysis on the decomposed wavelet signal.

According to one side, the feature extractor may extract a first feature corresponding to at least one of wavelet energy and Shannon wavelet entropy through wavelet domain analysis.

According to one side, the feature extracting unit selects at least one of mean, standard deviation, variance, kurtosis, and skewness of the decomposed wavelet signal through wavelet statistical analysis. A first feature corresponding to the statistical data may be extracted.

According to one side, the feature extraction unit may extract a second feature output from a feature selection algorithm that receives the first feature as an input as feature data.

According to one side, the feature selection algorithm may output the second feature by performing chi-square tests and correlation analysis on the first feature.

According to one side, the failure detection unit may train a machine learning-based classifier based on the extracted feature data.

According to one side, the data collection unit receives current data corresponding to a preset time from at least one motor, and the failure detection unit classifies the defect corresponding to the current data corresponding to the preset time using a machine learning-based classifier, , it is possible to detect the failure of the robot based on the classification result of the defect.

According to one side, machine learning-based classifiers include linear discriminant analysis (LDA), fine tree (FT), naive bayes (NB), and support vector machine (SVM). It may be a classifier based on at least one of

A robot failure detection method according to an embodiment of the present invention comprises the steps of: collecting current data corresponding to at least one of a mechanical defect, an electronic defect, and an electrical defect from at least one motor installed in a bot in a data collecting unit; performing pre-processing of applying a dimension reduction technique to the current data collected by the data pre-processing unit, and extracting feature data according to at least one defect based on wavelet analysis of the pre-processed data in the feature extracting unit It may include detecting a failure of the robot based on the feature data extracted by the detection unit.

According to one side, the extracting of the feature data may include performing at least one of wavelet domain analysis and wavelet statistical analysis on the decomposed wavelet signal to extract the first feature.

According to one side, the extracting of the feature data may include extracting the second feature output from the feature selection algorithm that receives the first feature as the feature data.

According to an embodiment, the present invention can easily detect and diagnose a mechanical failure, an electronic failure, and an electrical failure of a robot through motor current signature analysis (MCSA) and machine learning.

According to one embodiment, the present invention can accurately classify defects according to mechanical failure, electronic failure, and electrical failure of the robot using only the three-phase current of the electric motor.

According to an embodiment of the present invention, when real-time implementation, calculation time can be shortened to have a fast response speed, and sensor components can be greatly reduced, thereby reducing the cost of detecting and diagnosing a failure.

1 is a view for explaining an apparatus for detecting a failure of a robot according to an embodiment.
2 is a view for explaining an example of collecting current data in the device for detecting a failure of a robot according to an embodiment.
3A to 3B are diagrams for explaining an example of pre-processing current data in an apparatus for detecting a failure of a robot according to an embodiment.
4 is a view for explaining an example of decomposing preprocessed data into a wavelet signal in the apparatus for detecting a failure of a robot according to an embodiment.
5 is a view for explaining an example of extracting feature data from the failure detection apparatus of the robot according to an embodiment.
6A to 6D are diagrams for explaining an example of a machine learning-based classifier of an apparatus for detecting a failure of a robot according to an embodiment.
7 is a view for explaining a method for detecting a failure of a robot according to an embodiment.

Hereinafter, various embodiments of the present document will be described with reference to the accompanying drawings.

Examples and terms used therein are not intended to limit the technology described in this document to specific embodiments, and should be understood to include various modifications, equivalents, and/or substitutions of the embodiments.

In the following, when it is determined that a detailed description of a known function or configuration related to various embodiments may unnecessarily obscure the gist of the present invention, a detailed description thereof will be omitted.

In addition, terms to be described later are terms defined in consideration of functions in various embodiments, which may vary according to intentions or customs of users and operators. Therefore, the definition should be made based on the content throughout this specification.

In connection with the description of the drawings, like reference numerals may be used for like components.

The singular expression may include the plural expression unless the context clearly dictates otherwise.

In this document, expressions such as “A or B” or “at least one of A and/or B” may include all possible combinations of items listed together.

Expressions such as "first," "second," "first," or "second," can modify the corresponding elements, regardless of order or importance, and to distinguish one element from another element. It is used only and does not limit the corresponding components.

When an (eg, first) component is referred to as being “(functionally or communicatively) connected” or “connected” to another (eg, second) component, that component is It may be directly connected to the element, or may be connected through another element (eg, a third element).

As used herein, "configured to (or configured to)" according to the context, for example, hardware or software "suitable for," "having the ability to," "modified to ," "made to," "capable of," or "designed to," may be used interchangeably.

In some contexts, the expression “a device configured to” may mean that the device is “capable of” with other devices or components.

For example, the phrase “a processor configured (or configured to perform) A, B, and C” refers to a dedicated processor (eg, an embedded processor) for performing the operations, or by executing one or more software programs stored in a memory device. , may refer to a general-purpose processor (eg, a CPU or an application processor) capable of performing corresponding operations.

Also, the term 'or' means 'inclusive or' rather than 'exclusive or'.

That is, unless stated otherwise or clear from context, the expression 'x employs a or b' means any one of natural inclusive permutations.

In the above-described specific embodiments, elements included in the invention are expressed in the singular or plural according to the specific embodiments presented.

However, the singular or plural expression is appropriately selected for the situation presented for convenience of description, and the above-described embodiments are not limited to the singular or plural component, and even if the component is expressed in plural, it is composed of a singular or , even a component expressed in a singular may be composed of a plural.

On the other hand, although specific embodiments have been described in the description of the invention, various modifications are possible without departing from the scope of the technical idea contained in the various embodiments.

Therefore, the scope of the present invention should not be limited to the described embodiments, but should be defined by the claims described below as well as the claims and equivalents.

1 is a view for explaining an apparatus for detecting a failure of a robot according to an embodiment.

Referring to FIG. 1 , the failure detection apparatus 100 according to an embodiment easily detects and detects mechanical failures, electronic failures and electrical failures of the robot through motor current signature analysis (MCSA) and machine learning. can be diagnosed

In addition, the failure detection apparatus 100 may accurately classify defects according to mechanical failures, electronic failures, and electrical failures of the robot using only the three-phase current of the electric motor.

In addition, the failure detection apparatus 100 may have a fast response speed due to a shortened calculation time when implemented in real time, and may reduce the cost of failure detection and diagnosis by greatly reducing sensor components.

Specifically, the failure detection device 100 collects and analyzes current data from the electric motor at any time to more accurately detect mechanical failures, electronic failures, and defects (mechanical defects, electronic defects, and electrical defects) according to the robot's mechanical failure, electronic failure, and electrical failure. It can be classified and classified.

Electrical faults are caused by short circuits between three-phase power lines, short circuits between the inner coils of the stator windings, and short circuits between windings of the same coil, also called inter-turn short circuit failures.

Electrical failures can be classified into electronic failures caused by defects in the power converter and inverter that control the electric motor. Occurs due to a switching fault that occurs in

On the other hand, mechanical defects include misalignment of teeth, cracked teeth, cracked or bent rotor and shaft, misalignment and unbalanced coupling of rolling element bearing (REB), eccentric bearing fault of RV reducer, and It may include an aging fault of the RV reducer, and the like.

In general, electrical and electronic faults are mostly detected through motor current signature analysis (MCSA), which analyzes the current flowing in the electric motor, and mechanical faults are detected through vibration signal analysis, lubrication oil analysis (ferrography), and acoustic emission analysis. do.

However, differences in analysis methods due to various sensors and data related to the above-described monitoring, defect detection, and diagnosis process make PHM a more complex problem.

Accordingly, the failure detection device 100 uses the electromechanical relationship to analyze the current data of the electric motor connected to the mechanical part by only analyzing the motor current signature, and a technique for more accurately classifying and classifying mechanical defects, electronic defects and electrical defects. to provide.

To this end, the failure detection apparatus 100 may include a data collection unit 110 , a data preprocessor 120 , a feature extraction unit 130 , and a failure detection unit 140 .

The data collection unit 110 according to an embodiment may collect current data corresponding to at least one of a mechanical defect, an electronic defect, and an electrical defect from at least one motor installed in the robot. For example, the motor may be an electric motor installed in an industrial robot.

According to one side, the data collection unit 110 may collect three-phase current data I _abc of at least one motor derived through motor current signature analysis (MCSA).

The data preprocessor 120 according to an embodiment may perform preprocessing for applying a dimension reduction technique to the collected current data.

According to one side, the data preprocessor 120 may reduce the three-phase current data I _abc to the two-phase current data I _dq by using DQ0 transformation.

The feature extractor 130 according to an embodiment may extract feature data according to at least one defect based on wavelet analysis of the preprocessed data.

According to one side, the feature extractor 130 may decompose preprocessed data into wavelet signals through time-domain analysis based on discrete wavelet transform (DWT).

According to one side, the feature extraction unit 130 may extract the first feature by performing at least one of wavelet domain analysis and wavelet statistical analysis on the decomposed wavelet signal.

Specifically, the feature extractor 130 may extract a first feature corresponding to at least one of wavelet energy and Shannon wavelet entropy through wavelet domain analysis.

In addition, the feature extractor 130 performs at least one of a mean, standard deviation, variance, kurtosis, and skewness of the decomposed wavelet signal through wavelet statistical analysis. A first feature corresponding to one piece of statistical data may be extracted.

According to one side, the feature extraction unit 130 may extract the second feature output from the feature selection algorithm that receives the first feature as the feature data.

For example, the feature selection algorithm may perform chi-square tests and correlation analysis on the first feature to output the second feature.

The failure detection unit 140 according to an embodiment may detect a failure of the robot based on the extracted feature data.

According to one side, the failure detection unit 140 may learn a machine learning-based classifier based on the extracted feature data.

According to one side, the data collection unit 110 may receive current data corresponding to a preset time from at least one motor, and the failure detection unit 140 may receive current data corresponding to a preset time using a machine learning-based classifier. A defect corresponding to the current data may be classified, and a failure of the robot may be detected based on the classification result of the defect.

For example, the current data corresponding to the preset time may be real-time current data corresponding to a preset time range based on the current time.

In addition, the machine learning-based classifier is at least one of linear discriminant analysis (LDA), fine tree (FT), naive bayes (NB), and support vector machine (SVM). It may be a classifier based on one.

Preferably, the classifier based on machine learning may be a classifier based on linear discriminant analysis (LDA) or support vector machine (SVM).

2 is a view for explaining an example of collecting current data in the device for detecting a failure of a robot according to an embodiment.

Referring to FIG. 2, (a) to (c) of FIG. 2 are three-phase current data (phase A, B) collected through motor current signature analysis (MCSA) by the data collection unit of the failure detection device according to an embodiment. , C)(I _abc ) is shown.

Specifically, the data collection unit may collect motor current data for each of the three phases using a plurality of current sensors installed at positions corresponding to each of the three phases of the electric motor.

For example, the data collection unit may collect 18 current data from 18 current sensors installed for a total of 6 electric motors corresponding to each phase of the electric motor.

3A to 3B are diagrams for explaining an example of pre-processing current data in an apparatus for detecting a failure of a robot according to an embodiment.

3A to 3B , reference numeral 310 denotes two-phase current data (I) by applying DQ0 transformation to three-phase current data (I _abc ) in the data preprocessing unit of the failure detection apparatus according to an embodiment. _dq ) is shown as a 2D image, and reference numeral 320 is an image for explaining the process of dimensionally reducing the three-phase current data (I _abc ) to the two-phase current data (I _dq ) through the DQ0 transformation. show

Referring to reference numeral 310, the data preprocessor may DQ0-convert the three-phase current data I _abc shown in FIGS. 2A to 2C to output the two-phase current data I _dq .

Referring to reference numeral 320 , the data preprocessor may output the two-phase current data I _dq by using at least one of a cosine-based DQ0 transformation and a sine-based DQ0 transformation.

The cosine-based DQ0 transform is aligned to the A axis, where the d axis is the reference axis, and the sine-based DQ0 transform is aligned to the A axis, where the q axis is the reference axis. have.

The cosine-based DQ0 transform can output the result (d = 0, q = -1, zero = 0) by aligning the rotating DQ frame to the A axis at time t = 0, while the sine-based DQ0 transform can The result (d = 1, q = 0, zero = 0) can be output by aligning the rotating DQ frame at time t = 0 to the axis rotated 90 degrees behind the A axis.

The data preprocessor according to an embodiment may convert the three-phase current data I _abc into the two-phase current data I _dq by applying Equation 1 below for performing the sine-based DQ0 conversion.

[Equation 1]

4 is a view for explaining an example of decomposing preprocessed data into a wavelet signal in the apparatus for detecting a failure of a robot according to an embodiment.

Referring to FIG. 4 , reference numeral 400 denotes two-phase data preprocessed by the feature extraction unit of the failure detection device according to an embodiment, that is, two-phase current data (I _dq ) through DWT-based time-frequency domain analysis. An example of decomposing the current data I _dq into a wavelet signal is shown.

Analysis of the two-phase current data I _dq may be performed through at least one of a time domain analysis, a frequency domain analysis, and a time-frequency domain analysis.

In the time domain analysis, statistical features that summarize useful information in the time domain can be extracted from the signal, and the frequency domain analysis is known as an analysis advantageous for distinguishing defects with specific characteristics. However, time domain analysis and frequency domain analysis have limitations in analyzing abnormal data (eg, mechanical defects).

Accordingly, in order to overcome the above-described limitation, the feature extraction unit according to an embodiment uses a time-frequency analysis in which the frequency domain and the time domain are mixed, but uses a DWT that has an advantage in temporal resolution for capturing both frequency and location information. can be used to analyze the two-phase current data (I _dq ).

Referring to reference numeral 400, the feature extraction unit according to an embodiment extracts the two-phase current data (I _dq ) as shown in Equation 2 below: an approximation component A _m [n] and a detail component D _m [ n] can be decomposed.

[Equation 2]

Specifically, the feature extractor may convolve the two-phase current data I _d [n] and I _q [n] using QMF (quadrature mirror filter) h and q, and the data sequence according to this may be level 1 can be decomposed into a first approximate component A ₁ [n] and a first sub-component D ₁ [n]. Next, in level 2, the first approximate component A ₁ [n] can be decomposed into a second approximate component A ₂ [n] and a second sub-component D ₂ [n].

Preferably, the feature extraction unit repeats the decomposition up to level 6, so that the fifth approximate component A ₅ [n] can be decomposed into a sixth approximate component A ₆ [n] and a sixth sub-component D ₆ [n].

5 is a view for explaining an example of extracting feature data from the failure detection apparatus of the robot according to an embodiment.

Referring to FIG. 5 , reference numeral 500 illustrates an implementation example of a feature extracting unit for extracting features in the failure detection apparatus according to an embodiment.

The feature extractor 500 may extract a plurality of features from the decomposed wavelet signal when DWT is performed on the preprocessed data.

However, when the feature extracted through the DWT in the feature extraction unit 500 is an inappropriate feature, if an inappropriate feature is learned in the learning process of the machine learning-based classifier performed by the failure detection unit later, the accuracy of the defect classification in the classifier , which can reduce the reliability of the classification.

Accordingly, the feature extraction unit 500 according to an embodiment may perform a process of increasing the reliability of classification by extracting and selecting an appropriate defect discrimination feature from among the features extracted through the DWT.

To this end, the feature extractor 500 may include a wavelet domain analyzer 510 , a wavelet statistical analyzer 520 , and a feature selector 530 .

The wavelet domain analyzer 510 may extract a first feature corresponding to at least one of wavelet energy and Shannon wavelet entropy through wavelet domain analysis of the decomposed wavelet signal.

Specifically, the wavelet domain analyzer 510 may extract the first feature corresponding to the wavelet energy and the first feature corresponding to the Shannon wavelet entropy as in Equation 3 from the decomposed wavelet signal.

[Equation 3]

Here, A _i and D _i each represent an approximate component and a detailed component of the wavelet signal decomposed at the i-th level, s represents an input signal, and si represents an orthonormal basis of the input signal s.

For example, the wavelet domain analyzer 510 through Equation 3, the two-phase current data component (I _d and Twenty-four first features as shown in Table 1 below can be extracted from each subcomponent (D ₁ to D ₆ ) of I _q ).

The wavelet statistical analysis unit 520 performs a wavelet statistical analysis on the decomposed wavelet signal, and the mean, standard deviation, variance, kurtosis, and skewness of the decomposed wavelet signal. A first feature corresponding to at least one piece of statistical data among (skewness) may be extracted.

Specifically, the wavelet statistical analysis unit 520 calculates the mean, standard deviation, variance, kurtosis, and skewness of the decomposed wavelet signal as shown in Equation 4 below. A corresponding first feature may be extracted.

[Equation 4]

Here, n represents one observation and N represents the total number of observations.

For example, the wavelet statistical analysis unit 520 through Equation 4, the two-phase current data component (I _d and From each subcomponent (D ₁ to D ₆ ) of I _q ), 60 first features as shown in Table 2 below can be extracted.

Meanwhile, the feature selector 530 extracts the second feature output from the feature selection algorithm that receives the first feature output through the wavelet domain analyzer 510 and the wavelet statistical analyzer 520 as input as feature data. can do.

According to one side, the feature selector 530 may output the second feature by performing chi-square tests and correlation analysis.

For example, when the number of first features received from the wavelet domain analyzer 510 is 24 and the number of features received through the wavelet statistical analyzer 520 is 60, the feature selector 530 is a chi-square. Twenty features from among 84 first features may be selected through the test, and 10 features may be selected from among the 20 features selected through correlation analysis as second features.

That is, the feature extracting unit 500 according to an embodiment performs current data through DWT-based time-frequency domain analysis, wavelet domain analysis and wavelet statistical analysis based on the time-frequency domain analysis result, chi-square test, and correlation analysis. More significant features can be extracted from

6A to 6D are diagrams for explaining an example of a machine learning-based classifier of an apparatus for detecting a failure of a robot according to an embodiment.

6A to 6D , reference numerals 610 to 640 denote non-defective classifiers using machine learning-based classifiers learned through feature data received from the feature extraction unit in the failure detection unit of the failure detection apparatus according to an embodiment. _The results (output class) of classifying current data (target class) corresponding to (normal state, normal), the eccentric bearing fault (faulty) of the RV reducer, and the aging fault (faulty aged) of the RV reducer are shown.

Specifically, reference numeral 610 denotes a classification result of a linear discriminant analysis (LDA)-based classifier, reference numeral 620 denotes a classification result of a pine tree (FT)-based classifier, and reference numeral 630 denotes a naive Bayes ( NB) shows the classification result of the classifier, and reference numeral 640 denotes the classification result of the support vector machine (SVM)-based classifier.

According to reference numerals 610 to 640, the average accuracy of the linear discriminant analysis (LDA)-based classifier is 96.7%, the average accuracy of the pine tree (FT)-based classifier is 70%, and the average of the naive Bayes (NB)-based classifier is The accuracy was 33.3%, and the average accuracy of the support vector machine (SVM)-based classifier was 93.3%.

Meanwhile, the results of checking the accuracy by varying the training data of each classifier are shown in Table 3 below.

Here, case1 represents the average accuracy of the machine learning-based classifier trained with 24 features derived through wavelet domain analysis, and case2 represents the average of the machine learning-based classifier learned with 60 features derived through wavelet statistical analysis. The accuracy is shown, case 3 represents the average accuracy of the machine learning-based classifier trained with 20 features derived through the chi-square test, and case 4 represents the machine learning-based classifier learned with 20 features derived through correlation analysis. represents the average accuracy of

In addition, case5 represents the average accuracy of the machine learning-based classifier corresponding to the feature data derived through wavelet domain analysis, wavelet statistical analysis, chi-square test, and correlation analysis, that is, reference numerals 610 to 640 .

According to Table 3, it was found that the average accuracy of the classifier learned from the feature data according to an embodiment among cases 1 to 5 was 73.3%, which was the highest.

In particular, in case 5, the linear discriminant analysis (LDA)-based classifier (average accuracy 96.7%) and the support vector machine (SVM)-based classifier (average accuracy 93.3%) showed the highest accuracy.

7 is a view for explaining a method for detecting a failure of a robot according to an embodiment.

In other words, FIG. 7 is a view for explaining a method of operating the apparatus for detecting a failure of a robot according to an embodiment described with reference to FIGS. 1 to 6D , and among the contents described with reference to FIG. 7 later, through FIGS. 1 to 6D . A description that overlaps with the description will be omitted.

Referring to FIG. 7 , in step 710, the method for detecting a failure of a robot according to an embodiment includes current data corresponding to at least one of a mechanical defect, an electronic defect, and an electrical defect from at least one motor installed in the robot in the data collection unit. can be collected.

Next, in step 720 , the robot failure detection method according to an embodiment may perform a pre-processing of applying a dimension reduction technique to the current data collected by the data pre-processing unit.

Next, in step 730 , the robot failure detection method according to an embodiment may extract feature data according to at least one defect based on wavelet analysis of data preprocessed by the feature extraction unit.

According to one side, in step 730, the robot failure detection method according to an embodiment may perform at least one of wavelet domain analysis and wavelet statistical analysis on the decomposed wavelet signal to extract the first feature.

Also, in step 730 , the robot failure detection method according to an embodiment may extract a second feature output from a feature selection algorithm that receives the first feature as an input as feature data.

Next, in step 740 , the robot failure detection method according to an embodiment may detect a failure of the robot based on the feature data extracted by the failure detection unit.

As a result, using the present invention, it is possible to easily detect and diagnose a mechanical failure, an electronic failure, and an electrical failure of a robot through motor current signature analysis (MCSA) and machine learning.

In addition, the present invention can accurately classify defects according to mechanical failure, electronic failure, and electrical failure of the robot using only the three-phase current of the electric motor.

In addition, the present invention can have a fast response speed by shortening the calculation time in real-time implementation, and can reduce the cost of detecting a fault and diagnosing a failure by greatly reducing sensor components.

As described above, although the embodiments have been described with reference to the limited drawings, various modifications and variations are possible from the above description by those of ordinary skill in the art. For example, the described techniques are performed in an order different from the described method, and/or the described components of the system, structure, apparatus, circuit, etc. are combined or combined in a different form than the described method, or other components Or substituted or substituted by equivalents may achieve an appropriate result.

Therefore, other implementations, other embodiments, and equivalents to the claims are also within the scope of the following claims.

100: failure detection device 110: data collection unit
120: data preprocessor 130: feature extraction unit
140: fault detection unit

Claims (15)

a data collection unit for collecting current data corresponding to at least one of a mechanical defect, an electronic defect, and an electrical defect from at least one motor installed in the robot;
a data preprocessing unit performing preprocessing for applying a dimension reduction technique to the collected current data;
The preprocessed data is decomposed into a wavelet signal through time-domain analysis based on Discrete Wavelet Transform (DWT), and wavelet domain analysis and wavelet statistical analysis are performed on the decomposed wavelet signal. Extracts one feature, and sets a second feature output from a feature selection algorithm that performs chi-square tests and correlation analysis on the first feature according to the at least one defect. a feature extraction unit for extracting feature data; and
A failure detection unit that detects a failure of the robot based on the extracted feature data
A robot failure detection device comprising a. According to claim 1,
The data collection unit,
Collecting three-phase current data of the at least one motor derived through motor current signature analysis (MCSA)
Robot fault detection device. 3. The method of claim 2,
The data preprocessor,
The three-phase current data is dimensionally reduced to two-phase current data using DQ0 transformation.
Robot fault detection device. delete delete According to claim 1,
The feature extraction unit,
extracting the first feature corresponding to at least one of wavelet energy and Shannon wavelet entropy through the wavelet domain analysis;
Robot fault detection device. According to claim 1,
The feature extraction unit,
Through the wavelet statistical analysis, at least one of a mean, standard deviation, variance, kurtosis, and skewness of the decomposed wavelet signal corresponding to statistical data extracting the first feature
Robot fault detection device. delete delete According to claim 1,
The fault detection unit,
Learning a machine learning-based classifier based on the extracted feature data
Robot fault detection device. 11. The method of claim 10,
The data collection unit,
receiving current data corresponding to a preset time from the at least one motor;
The fault detection unit,
Classifying a defect corresponding to the current data corresponding to the preset time using the machine learning-based classifier, and detecting a failure of the robot based on the classification result of the defect
Robot fault detection device. 11. The method of claim 10,
The machine learning-based classifier is
a classifier based on at least one of linear discriminant analysis (LDA), fine tree (FT), naive bayes (NB), and support vector machine (SVM).
Robot fault detection device. collecting, in the data collection unit, current data corresponding to at least one of a mechanical defect, an electronic defect, and an electrical defect from at least one motor installed in the robot;
performing, in a data preprocessing unit, preprocessing for applying a dimension reduction technique to the collected current data;
In the feature extraction unit, the preprocessed data is decomposed into a wavelet signal through time-domain analysis based on Discrete Wavelet Transform (DWT), and wavelet domain analysis and wavelet statistics for the decomposed wavelet signal are performed. performing analysis to extract a first feature, and using a second feature output from a feature selection algorithm that performs chi-square tests and correlation analysis on the first feature at the at least extracting feature data according to one defect; and
Detecting, in a failure detection unit, a failure of the robot based on the extracted feature data
A robot failure detection method comprising a. delete delete

Images