# US 11,307,570 B2 · Machine learning based predictive maintenance of equipment

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 11,307,570 B2
- Document ID: US11307570B2
- Status: Granted
- Kind: B2
- Application number: 16/887,957
- Filed: 2020-05-29
- Published: 2020-12-03
- Granted: 2022-04-19
- Earliest claimed: 2019-05-31
- Assignees: Panasonic Intellectual Property Management Co., Ltd.
- Applicants: Panasonic Intellectual Property Management Co., Ltd.
- Inventors: Hai Anh Trinh; Christopher T. Nguyen; The Vinh Luong; Taejin Chun
- Web page: https://patentgrep.com/patent/US11307570B2

## CPC

- G05B 23/0283: {Predictive maintenance, e.g. involving the monitoring of a system and, based on the monitoring results, taking decisions on the maintenance schedule of the monitored system; Estimating remaining useful life [RUL] (preventive maintenance, i.e. planning maintenance according to the available resources without monitoring the system G06Q10/06)}
- G05B 23/024: {Quantitative history assessment, e.g. mathematical relationships between available data; Functions therefor; Principal component analysis [PCA]; Partial least square [PLS]; Statistical classifiers, e.g. Bayesian networks, linear regression or correlation analysis; Neural networks}
- G06F 17/18: for evaluating statistical data {, e.g. average values, frequency distributions, probability functions, regression analysis (forecasting specially adapted for a specific administrative, business or logistic context G06Q10/04)}
- G06N 3/045: Combinations of networks
- G06N 3/0455: Auto-encoder networks; Encoder-decoder networks
- G06N 3/047: Probabilistic or stochastic networks
- G06N 3/088: Non-supervised learning, e.g. competitive learning
- G06N 3/0895: Weakly supervised learning, e.g. semi-supervised or self-supervised learning
- G06N 3/09: Supervised learning
- G06N 20/00: Machine learning
- G06N 3/044: Recurrent networks, e.g. Hopfield networks
- G06N 3/084: Backpropagation, e.g. using gradient descent
- G06N 5/01: Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound
- G06N 20/10: using kernel methods, e.g. support vector machines [SVM]
- G06N 20/20: Ensemble learning
- G05B: CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS (systems for controlling or regulating non-electric variables G05D; systems for regulating electric or magnetic variables G05F; control devices or systems insofar as characterised by mechanical features only G05G)
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

A predictive maintenance server receives data from sensors of equipment. The server uses one or more machine learning models to assign an anomaly score. Responsive to the anomaly score exceeding a threshold value, the server may issue an alert. The machine learning model may be supervised or unsupervised. In one embodiment, the machine learning model use several sensor channels to predict the values of one or more vitals of the equipment and compare the predicted values to the actual measured values of the vitals. The server may assign an anomaly score based on the differences between the predicted values and the measured values. In one embodiment, the machine learning model may be an autoencoder that generates a distribution of the measurement values to determine the likelihood of observing the actual measured values in a normal operation. In one embodiment, the server may use a histogram approach to predict anomaly.

## Claims

### Claim 1

A computer-implemented method comprising:
receiving a set of sensor data generated from sensors associated with equipment, one of the sensors being a target sensor, the set of sensor data comprising measured values of the target sensor;
selecting a subset of sensor data, the subset of sensor data comprising data generated from the sensors and excluding the measured values of the target sensor;
inputting the subset of sensor data into a machine learning model to generate predicted sensor values of the target sensor;
determining differences between the predicted sensor values of the target sensor and the measured values of the target sensor;
normalizing the differences using a reference metric of the machine learning model, the reference metric corresponding to training error of the machine learning model in predicting values of the target sensor;
generating an anomaly score for the equipment based on the differences; and
generating, based on anomaly score, an alert for the equipment.

### Claim 2 (depends on 1)

The computer-implemented method of claim 1, wherein the target sensor is a first target sensor, the sensors comprise a second target sensor, wherein the subset of sensor data is a first subset, and the computer-implemented method further comprises:
selecting a second subset of sensor data comprising data generated from the sensors including the first target sensor and excluding the second target sensor;
inputting the second subset of sensor data into a second machine learning model to generate second predicted sensor values of the second target sensor;
determining second differences between the second predicted values of the second target sensor and second measured values of sensor data of the second target sensor; and
wherein the anomaly score is generated further based on the second differences.

### Claim 3 (depends on 1)

The computer-implemented method of claim 1, wherein training of the machine learning model comprises:
receiving a set of training sensor data comprising historical data of the sensors, the set of training sensor data comprising historical measured values of the target sensor;
determining a subset of training sensor data comprising the historical data generated from the sensors excluding the target sensor;
inputting the subset of training sensor data to the machine learning model to generate outputs of the machine learning models; and
adjusting weights of the machine learning models to reduce training errors of the outputs compared to the historical measured values of the target sensor.

### Claim 4 (depends on 3)

The computer-implemented method of claim 3, wherein a reference metric is an average of differences between the outputs and the historical measured values when the machine learning model is trained.

### Claim 5 (depends on 4)

The computer-implemented method of claim 4, wherein the average of differences between the outputs and the historical measured values is determined using the historical measured values that were generated over a static period of time.

### Claim 6 (depends on 4)

The computer-implemented method of claim 4, wherein the average of differences between the outputs and the historical measured values is determined using the historical measured values that were generated over a sliding window of time preceding a time in which the set of sensor data is generated.

### Claim 7

A computer-implemented method comprising:
receiving a set of sensor data generated from sensors associated with equipment, one of the sensors being a target sensor, the set of sensor data comprising measured values of the target sensor;
selecting a subset of sensor data, the subset of sensor data comprising data generated from the sensors and excluding the measured values of the target sensor;
inputting the subset of sensor data into a machine learning model to generate predicted sensor values of the target sensor;
determining differences between the predicted sensor values of the target sensor and the measured values of the target sensor, wherein training of the machine learning model comprises:
receiving a set of training sensor data comprising historical data of the sensors, the set of training sensor data comprising historical measured values of the target sensor,
determining a subset of training sensor data comprising the historical data generated from the sensors excluding the target sensor,
inputting the subset of training sensor data to the machine learning model to generate outputs of the machine learning models, and
adjusting weights of the machine learning models to reduce training errors of the outputs compared to the historical measured values of the target sensor;
generating an anomaly score for the equipment based on the differences; and
generating, based on anomaly score, an alert for the equipment.

### Claim 8 (depends on 7)

The computer-implemented method of claim 7, wherein the target sensor is a first target sensor, the sensors comprise a second target sensor, wherein the subset of sensor data is a first subset, and the computer-implemented method further comprises:
selecting a second subset of sensor data comprising data generated from the sensors including the first target sensor and excluding the second target sensor;
inputting the second subset of sensor data into a second machine learning model to generate second predicted sensor values of the second target sensor;
determining second differences between the second predicted values of the second target sensor and second measured values of sensor data of the second target sensor; and
wherein the anomaly score is generated further based on the second differences.

### Claim 9 (depends on 7)

The computer-implemented method of claim 7, wherein a reference metric is an average of differences between the outputs and the historical measured values when the machine learning model is trained.

### Claim 10 (depends on 9)

The computer-implemented method of claim 9, wherein the average of differences between the outputs and the historical measured values is determined using the historical measured values that were generated over a static period of time.

### Claim 11 (depends on 9)

The computer-implemented method of claim 9, wherein the average of differences between the outputs and the historical measured values is determined using the historical measured values that were generated over a sliding window of time preceding a time in which the set of sensor data is generated.

### Claim 12

A non-transitory computer-readable medium configured to store computer code comprising instructions, the instructions, when executed by a processor, cause the processor to perform steps comprising:
receiving a set of sensor data generated from sensors associated with equipment, one of the sensors being a target sensor, the set of sensor data comprising measured values of the target sensor;
selecting a subset of sensor data, the subset of sensor data comprising data generated from the sensors and excluding the measured values of the target sensor;
inputting the subset of sensor data into a machine learning model to generate predicted sensor values of the target sensor;
determining differences between the predicted sensor values of the target sensor and the measured values of the target sensor;
normalizing the differences using a reference metric of the machine learning model, the reference metric corresponding to training error of the machine learning model in predicting values of the target sensor;
generating an anomaly score for the equipment based on the differences; and
generating, based on anomaly score, an alert for the equipment.

### Claim 13 (depends on 12)

The non-transitory computer-readable medium of claim 12, wherein the target sensor is a first target sensor, the sensors comprise a second target sensor, wherein the subset of sensor data is a first subset, and the steps further comprises:
selecting a second subset of sensor data comprising data generated from the sensors including the first target sensor and excluding the second target sensor;
inputting the second subset of sensor data into a second machine learning model to generate second predicted sensor values of the second target sensor;
determining second differences between the second predicted values of the second target sensor and second measured values of sensor data of the second target sensor; and
wherein the anomaly score is generated further based on the second differences.

### Claim 14 (depends on 12)

The non-transitory computer-readable medium of claim 12, wherein training of the machine learning model comprises:
receiving a set of training sensor data comprising historical data of the sensors, the set of training sensor data comprising historical measured values of the target sensor;
determining a subset of training sensor data comprising the historical data generated from the sensors excluding the target sensor;
inputting the subset of training sensor data to the machine learning model to generate outputs of the machine learning models; and
adjusting weights of the machine learning models to reduce training errors of the outputs compared to the historical measured values of the target sensor.

### Claim 15 (depends on 14)

The non-transitory computer-readable medium of claim 14, wherein a reference metric is an average of differences between the outputs and the historical measured values when the machine learning model is trained.

### Claim 16 (depends on 15)

The non-transitory computer-readable medium of claim 15, wherein the average of differences between the outputs and the historical measured values is determined using the historical measured values that were generated over a static period of time.

### Claim 17 (depends on 15)

The non-transitory computer-readable medium of claim 15, wherein the average of differences between the outputs and the historical measured values is determined using the historical measured values that were generated over a sliding window of time preceding a time in which the set of sensor data is generated.