# US 2021/0334656 A1 · Computer-implemented method, computer program product and system for anomaly detection and/or predictive maintenance

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

- Number: US 2021/0334656 A1
- Document ID: US20210334656A1
- Status: Published application
- Kind: A1
- Application number: 17/273,709
- Filed: 2019-09-05
- Published: 2021-10-28
- Earliest claimed: 2018-09-05
- Assignees: Sartorius Stedim Data Analytics AB
- Applicants: Sartorius Stedim Data Analytics AB
- Inventors: Rickard Sjögren; Johan Trygg
- Web page: https://patentgrep.com/patent/US20210334656A1

## CPC

- 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}
- G06N 3/04: Architecture, e.g. interconnection topology
- 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)}
- G06F 18/213: Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F 18/22: Matching criteria, e.g. proximity measures
- G06N 3/006: based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
- G06N 3/044: Recurrent networks, e.g. Hopfield networks
- G06N 3/0442: characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
- G06N 3/045: Combinations of networks
- G06N 3/0455: Auto-encoder networks; Encoder-decoder networks
- G06N 3/0464: Convolutional networks [CNN, ConvNet]
- G06N 3/047: Probabilistic or stochastic networks
- G06N 3/0475: Generative networks
- G06N 3/048: Activation functions
- G06N 3/08: Learning methods
- G06N 3/082: modifying the architecture, e.g. adding, deleting or silencing nodes or connections
- G06N 3/084: Backpropagation, e.g. using gradient descent
- G06N 3/09: Supervised learning
- G06N 3/092: Reinforcement learning
- G06N 3/094: Adversarial learning
- G06N 5/045: Explanation of inference; Explainable artificial intelligence [XAI]; Interpretable artificial intelligence
- G06N 7/01: Probabilistic graphical models, e.g. probabilistic networks
- G06N 20/00: Machine learning
- G06N 20/10: using kernel methods, e.g. support vector machines [SVM]
- G06V 20/698: {Matching; Classification}
- G06N 3/088: Non-supervised learning, e.g. competitive 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)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING

## Abstract

An example method comprises receiving a new observation characterizing at least one parameter of an entity; inputting the new observation to a deep neural network having hidden layers; obtaining a second set of intermediate output values that are output from at least one of the hidden layers by inputting the received new observation to the deep neural network; mapping the second set of intermediate output values to a second set of projected values; determining whether or not the received new observation is an outlier with respect to the training dataset based on the latent variable model and the second set of projected values, calculating a prediction for the new observation; and determining a result indicative of the occurrence of at least one anomaly in the entity based on the prediction and the determination whether or not the new observation is an outlier.

## Claims

### Claim 1

A computer-implemented method for anomaly detection in an entity of interest comprising:
receiving a new observation said new observation characterizing at least one parameter of the entity;
inputting the new observation to a deep neural network (100), the deep neural network (100) having a plurality of hidden layers and being trained using a training data set that includes possible observations that can be input to the deep neural network (100);
obtaining a second set of intermediate output values that are output from at least one of the plurality of hidden layers of the deep neural network (100) by inputting the received new observation to the deep neural network (100);
mapping, using a latent variable model stored in a storage medium, the second set of intermediate output values to a second set of projected values;
determining whether or not the received new observation is an outlier with respect to the training dataset based on the latent variable model and the second set of projected values,
calculating, by the deep neural network (100), a prediction for the new observation; and
determining a result indicative of the occurrence of at least one anomaly in the entity based on the prediction and the determination whether or not the new observation is an outlier;
wherein the latent variable model stored in the storage medium is constructed by:
obtaining first sets of intermediate output values that are output from said one of the plurality of hidden layers of the deep neural network (100), each of the first sets of intermediate output values obtained by inputting a different one of the possible observations included in at least a part of the training dataset; and
constructing the latent variable model using the first sets of intermediate output values, the latent variable model providing a mapping of the first sets of intermediate output values to first sets of projected values in a sub-space of the latent variable model that has a dimension lower than a dimension of the sets of the intermediate outputs.

### Claim 2 (depends on 1)

A computer-implemented method according to claim 1, further comprising:
obtaining the deep neural network (100) for processing data and at least a part of a training dataset used for training the deep neural network (100), the deep neural network comprising a plurality of hidden layers and being trained by using the training dataset, the training dataset including a plurality of possible observations that can be input to the deep neural network;
obtaining first sets of intermediate output values that are output from at least one of the plurality of hidden layers, each of the first sets of intermediate output values obtained by inputting a different one of the possible observations included in said at least a part of the training dataset;
constructing or fitting a latent variable model using the first sets of intermediate output values, the latent variable model providing a mapping of the first sets of intermediate output values to first sets of projected values in a sub-space that has a dimension lower than a dimension of the sets of the intermediate outputs; and
storing the latent variable model and the first sets of projected values in a storage medium.

### Claim 3 (depends on 1)

The method according to claim 1, wherein said step of determining whether or not the received new observation is an outlier comprises:
determining a distance of the second set of projected values with respect to a distribution of the first sets of projected values; and
determining that the received new observation is an outlier with respect to the training dataset if the calculated distance is larger than a threshold value for the determined distance.

### Claim 4 (depends on 3)

The method according to claim 3, wherein the threshold value for the distance is determined based on distances, each of which being calculated for a different one of the first sets of projected values with respect to the distribution of the first sets of projected values.

### Claim 5 (depends on 3)

The method of claim 3, wherein the distance is one of a compound distance, residual sum of squares, Mahalanobis distance or Local Outlier Factor.

### Claim 6 (depends on 1)

The method according to claim 1, wherein said step of determining whether or not the received new observation is an outlier comprises:
determining an approximate set of intermediate output values corresponding to the second set of intermediate output values, using the latent variable model and the second set of projected values;
calculating a squared approximation residual for the second set of intermediate output values and the approximate set of intermediate output values; and
determining that the received observation is an outlier with respect to the training dataset if the calculated squared approximation residual is larger than a threshold value for the squared approximation residual.

### Claim 7 (depends on 6)

The method according to claim 6, wherein the threshold value for the squared approximation residual is determined based on squared approximation residuals, each of which is calculated for a different one of the first sets of intermediate output values and an approximate set of intermediate output values corresponding to said one of the first sets of intermediate output values.

### Claim 8 (depends on 1)

The method according to claim 1, wherein the steps of obtaining the first sets of intermediate output values and constructing the latent variable model are performed for two or more of the plurality of hidden layers;
wherein the steps of obtaining the second set of intermediate output values and mapping the second set of intermediate output values to the second set of projected values are performed concerning said two or more of the plurality of hidden layers; and
wherein, the step of determining whether or not the received new observation is an outlier is performed based on the latent variable model and the second sets of projected values obtained concerning said two or more of the plurality of hidden layers.

### Claim 9 (depends on 1)

The method according to claim 1, wherein the step of obtaining the intermediate output values that are output from at least one of the plurality of hidden layers of the deep neural network (100) comprises determining activations from the at least one of the plurality of hidden layers and optionally applying a global pooling on the determined activations.

### Claim 10 (depends on 1)

The method according to claim 1, wherein the latent variable model is constructed according to principal component analysis or using an autoencoder.

### Claim 11 (depends on 1)

A method according to claim 1, wherein the observations are obtained based on physical measurements by one or more sensors.

### Claim 12 (depends on 1)

A computer-implemented method for predictive maintenance of an entity, the method comprising:
obtaining a new observation based on sensor data from at least one sensor measurement performed on the entity;
determining a result indicative of the occurrence of at least one anomaly in the entity according to the method of claim 1; and
determining whether a corrective action should be taken and optionally, the type of the corrective action to be taken based on the result indicative of the occurrence of at least one anomaly.

### Claim 13 (depends on 1)

A computer program product comprising computer-readable instructions that, when loaded and run on a computer, cause the computer to perform the method according to claim 1.

### Claim 14 (depends on 1)

A system for anomaly detection or predictive maintenance of an entity, the system comprising:
a data acquisition unit for obtaining one or more observations characterizing at least one parameter of the entity;
a storage medium (12) storing a training dataset used for training a deep neural network (100), the deep neural network (100) comprising a plurality of hidden layers and being trained using the training dataset, the training dataset including possible observations that can be input to the deep neural network (100); and
a predictive analytics unit comprising at least one processor configured to perform the method of claim 1.

### Claim 15 (depends on 14)

The system according to claim 14 further comprising:
at least one sensor for measuring at least one parameter of the entity; and
a data communication unit for data transfer between the at least one sensor and the data acquisition unit and/or the at least one processor.