KR-102963567-B1 - APPARATUS, METHOD AND PROGRAM FOR ASSESSING BEARING CURRENT RISK USING MACHINE LEARNING

KR102963567B1KR 102963567 B1KR102963567 B1KR 102963567B1KR-102963567-B1

Abstract

It includes a data collection unit that collects raw data by measuring the motor-ground current flowing from the motor to the ground, a data preprocessing unit that distinguishes a stator-ground current band and an electric discharge bearing current band based on the frequency characteristics of sample data sampled from the raw data, and extracts time-domain statistical features from the stator-ground current band and the electric discharge bearing current band, a learning model generation unit that generates a learning model trained on the time-domain statistical features, and a risk evaluation unit that inputs the time-domain statistical features into the learning model to estimate the risk and remaining life of the motor.

Inventors

이영석

Assignees

멜콘 주식회사

Dates

Publication Date: 20260512
Application Date: 20250902

Claims (19)

In a bearing current risk assessment device using machine learning, A data acquisition unit that collects raw data by measuring the motor-ground current flowing from the motor to ground; A data preprocessing unit that distinguishes a stator-ground current band and an electric discharge bearing current band based on the frequency characteristics of sample data sampled from the raw data, and extracts time-domain statistical features independently for each of the stator-ground current band and the electric discharge bearing current band at the same sampling rate and the same time window; and A risk assessment unit that estimates the risk and remaining lifespan of the motor by inputting the time-domain statistical features into a pre-trained learning model; Includes, A bearing current risk assessment device using machine learning, wherein the time domain statistical features for the above remaining life estimation include the peak-to-peak of the stator-ground current band and the standard deviation and peak-to-peak of the electric discharge bearing current band.
In Article 1, The above data preprocessing unit A bearing current risk assessment device using machine learning, wherein the sampling frequency is set to be at least 10 times the predetermined frequency of interest, and the minimum standard for the number of samples is set to a sampling data length that includes a signal in which each phase of the inverter is switched twice and restored to its original state, thereby generating the sample data.
In Article 1, The above data preprocessing unit A spectrum is calculated by performing an FFT on the above sample data, and After obtaining the envelope of the above spectrum, the first local minimum is determined as the boundary frequency, and A bearing current risk assessment device using machine learning, which separates a signal into a stator-ground current band and an electric discharge bearing current band based on the above boundary frequency.
In Paragraph 3, The above data preprocessing unit A bearing current risk assessment device using machine learning, which extracts the stator-ground current band in the kHz band based on the above boundary frequency and extracts the electric discharge bearing current band in the MHz band.
In Article 1, The above data preprocessing unit A bearing current risk assessment device using machine learning, which calculates the above time-domain statistical features and forms them into a single normalized feature vector.
In Article 1, The above risk assessment department A bearing current risk assessment device using machine learning, wherein the time domain statistical features are input into the learning model to calculate the risk level from a risk score calculated from the ratio of a safe current value to a dangerous current value learned in advance.
In Article 6, The above risk assessment department A bearing current risk assessment device using machine learning, which calculates a ratio of lifespan for each predetermined risk interval based on an initial value of the lifespan in the normal operating state of the above-mentioned motor, and determines a lifespan reduction value by converting the calculated ratio of lifespan into a ratio of a predetermined sampling interval.
In Article 7, The above risk assessment department A bearing current risk assessment device using machine learning, wherein the remaining lifespan is updated by subtracting a lifespan reduction value calculated from a preset initial lifespan, and subsequently, the remaining lifespan data is generated by recursively recording the lifespan reduction value using the updated remaining lifespan as input.
In Article 8, The above risk assessment department A bearing current risk assessment device using machine learning, which estimates the remaining lifespan of the motor by calculating the point in time when the remaining lifespan becomes 0 in a graph through regression analysis of the remaining lifespan data.
In a bearing current risk assessment method using machine learning for a bearing current risk assessment device using machine learning, A step of collecting raw data by measuring the motor-ground current flowing from the motor to ground; A step of generating sample data by sampling the above raw data; A step of distinguishing the stator-ground current band and the electric discharge bearing current band based on the frequency characteristics of the above sample data; A step of independently extracting time-domain statistical features for each of the stator-ground current band and the electric discharge bearing current band at the same sampling rate and the same time window; and A step of estimating the risk and remaining lifespan of the motor by inputting the time-domain statistical features into a pre-trained learning model; Includes, A method for assessing bearing current risk using machine learning, wherein the time-domain statistical features for estimating the remaining life include the peak-to-peak of the stator-ground current band and the standard deviation and peak-to-peak of the electric discharge bearing current band.
In Article 10, The step of generating the above sample data A method for evaluating bearing current risk using machine learning, wherein the sampling frequency is set to be at least 10 times the predetermined frequency of interest, and the minimum standard for the number of samples is set to a sampling data length that includes a signal in which each phase of the inverter of the motor is switched twice and restored to its original state, thereby generating the sample data.
In Article 10, The above-mentioned distinguishing step A spectrum is calculated by performing an FFT on the above sample data, and After obtaining the envelope of the above spectrum, the first local minimum is determined as the boundary frequency, and A bearing current risk assessment method using machine learning, wherein the signal is separated into a stator-ground current band and an electric discharge bearing current band based on the above boundary frequency.
In Article 12, The above-mentioned distinguishing step A bearing current risk assessment method using machine learning, wherein the stator-ground current band is extracted in the kHz band based on the above boundary frequency, and the electric discharge bearing current band is extracted in the MHz band.
In Article 10, The above extraction step A bearing current risk assessment method using machine learning, wherein the above time-domain statistical features are calculated and constructed into a single normalized feature vector.
In Article 10, The above-mentioned estimation step A method for evaluating bearing current risk using machine learning, wherein the time domain statistical features are input into the learning model to calculate the risk level from a risk score calculated from the ratio of a safe current value to a dangerous current value learned in advance.
In Article 15, The above-mentioned estimation step A bearing current risk assessment method using machine learning, wherein the ratio of lifespan for each predetermined risk interval is calculated based on an initial value of the lifespan in the normal operating state of the above-mentioned motor, and the ratio of the calculated lifespan is converted into a ratio of a predetermined sampling interval to determine a lifespan reduction value.
In Article 16, The above-mentioned estimation step A method for assessing bearing current risk using machine learning, wherein the remaining lifespan is updated by subtracting a calculated lifespan reduction value from a preset initial lifespan, and then the remaining lifespan data is generated by recursively recording the lifespan reduction value using the updated remaining lifespan as input.
In Article 17, The above-mentioned estimation step A method for evaluating bearing current risk using machine learning, wherein the remaining life of the motor is estimated by calculating the point in time when the remaining life becomes 0 in the graph through regression analysis of the remaining life time data.
In a computer program stored on a computer-readable recording medium comprising instructions for providing a bearing current risk assessment method using machine learning, Collect raw data by measuring the motor-to-ground current flowing from the motor to ground, and Sample data is generated by sampling the above raw data, and Based on the frequency characteristics of the above sample data, the stator-ground current band and the electric discharge bearing current band are distinguished, and Time-domain statistical features are extracted independently for each of the above stator-ground current band and the above electric discharge bearing current band at the same sampling rate and the same time window, and It includes a sequence of commands that estimate the risk and remaining life of the motor by inputting the time-domain statistical features into a pre-trained learning model, and A computer program stored on a computer-readable recording medium, wherein the time-domain statistical features for the above remaining life estimation include the peak-to-peak of the stator-ground current band and the standard deviation and peak-to-peak of the electric discharge bearing current band.

Description

Apparatus, Method and Program for Assessing Bearing Current Risk Using Machine Learning The present invention relates to a bearing current risk assessment device, method, and program using machine learning. Electric motors, primarily used to generate rotational kinetic energy, are widely used across various industries due to their superior characteristics, ranging from the method of supplying electrical energy to their operational mechanisms, and their technological maturity has reached a very high level. Consequently, they demonstrate excellent performance not only in terms of efficiency but also in terms of lifespan. Despite significant advancements, there is a limit to the operating lifespan because they are devices that generate rotational kinetic energy. Bearings, which are essential components for motor operation and account for 40–50% of the causes of defects, are critical parts located between the stator and rotor that enable smooth rotational motion. Depending on the bearing, continuous operation is possible for tens of thousands to hundreds of thousands of hours if operating conditions are properly maintained. The major causes of bearing failure include poor lubrication, overload, installation errors, and electrical problems. Among electrical problems, some cause immediate failures, such as insulation breakdown, while others, like bearing current, lead to gradual failures. Bearing current, which was previously simply categorized under the phenomenon of shaft current, is now recognized as a cause of failure that passes through the bearings. It is often considered as a potential cause when an electric motor fails prematurely beyond its design life, even when there are no issues with its design or maintenance. Based on their characteristics, these bearing currents are classified into capacitive, electric discharge, circulating, and shaft-to-ground currents. Depending on the size and shape of the motor, these may or may not be the primary cause. Capacitive bearing current is generated by parasitic capacitance between the inner and outer rings of a bearing and is generally characterized by its small magnitude, which does not significantly affect bearing damage. Electric discharge bearing current is characterized by an increased potential on the rotor shaft, which damages the bearing's oil film and causes current to flow; it is classified as a major cause of bearing current problems, primarily in motors under 75kW. Circulating current is a bearing current that circulates along the path of shaft - half-load bearing - motor frame - load bearing, and is characterized by occurring mainly in motors over 75kW. Finally, shaft-ground bearing current is characterized by bearing damage caused by poor grounding between the motor and the ground, where the main path becomes shaft - bearing (motor or actuator) - ground. To measure these bearing currents, a current measuring device must be used along the bearing current path; however, in most cases, it is difficult to measure the current passing through the bearing. Nevertheless, the reason it is important to understand the bearing current is that the current may flow locally on the bearing surface, causing pitting by melting and re-solidification due to high temperatures, or cause fluting on the inner and outer rings of the bearing due to continuous discharge current, and can shorten the lifespan by causing deterioration (carbonization) of the lubricant due to discharge. For this reason, techniques for diagnosing failures and predicting lifespan have been developed through numerous studies in academia and industry. However, as variable speed drive systems (or inverters) are applied to control the high-efficiency rotational speed of induction motors, problems affecting the design life of bearings are increasing. One of the issues is that, regarding bearing current, research is currently being conducted on the relationship between inverter-driven induction motors and various bearing currents classified according to their generation causes and current paths, moving away from the concept of shaft current which was previously mainly generated in rotating machinery with large shafts, such as steam turbine generators, where static electricity is prone to generation and accumulation. FIG. 1 is a configuration diagram for explaining a bearing current risk assessment system using machine learning according to one embodiment of the present invention. FIG. 2 is a configuration diagram for explaining a bearing current risk assessment device using machine learning according to an embodiment of the present invention. FIG. 3 is a flowchart illustrating a bearing current risk assessment method using machine learning for estimating the remaining life of a motor according to an embodiment of the present invention. FIG. 4 is a flowchart illustrating a bearing current risk assessment method using machine learning for monitoring bearing current abnormal conditions according to an embodime