Sensor Time-Series Anomaly Detection Using Reconstruction Confidence

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional anomaly detection methods in semiconductor manufacturing processes are time-consuming, prone to inaccuracies, and fail to effectively handle multi-dimensional time series data with non-linearity and non-stationarity, lacking real-time automation and providing inadequate explanations for anomalies.

Innovation Solution

A neural network trained on prior sensor time series data is used to reconstruct sample data, determine anomaly scores based on reconstruction errors, and define confidence intervals for detecting anomalies in sensor time series data, enabling automatic and near real-time detection.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If manual anomaly detection by domain experts is used, then expertise-based accuracy is improved, but time consumption and process complexity increase significantly

Engineering Contradiction:

Improveanomaly detection accuracyVSAvoidtime consumption

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent replaces manual expert analysis with an automated machine learning system that uses neural networks and anomaly detection algorithms to analyze sensor time series data, eliminating the need for manual domain expert intervention while maintaining detection accuracy

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system enables self-service anomaly detection by automatically processing sensor data through trained models, generating anomaly scores and explanations without requiring external expert input, thereby reducing time consumption while preserving detection capability

Inventive Principle:

Principle #25Self-service

2Device complexity

If conventional anomaly detection methods are used, then simplicity is maintained, but they fail to handle multi-dimensional time series data with non-linearity and non-stationarity

Engineering Contradiction:

Improvedetection method simplicityVSAvoidanomaly detection reliability

Core Design Contradiction:

Device complexityVSReliability

Solution Approach 1:

The patent transforms the detection approach by changing parameters from simple statistical methods to advanced machine learning models that can handle non-linear and non-stationary data characteristics, improving reliability for complex manufacturing processes

Inventive Principle:

Principle #35Parameter changes

Solution Approach 2:

The system combines multiple computational components including neural networks, anomaly scoring mechanisms, and explanation generation modules to create a composite detection system that reliably handles multi-dimensional time series data while maintaining operational simplicity

Inventive Principle:

Principle #40Composite materials

3Measurement precision

If existing forecast models like LSTM or ARIMA are used, then confidence interval determination is improved, but they do not work when there is no seasonality or easy predictability in the data

Engineering Contradiction:

Improveconfidence interval determinationVSAvoidapplicability to various data patterns

Core Design Contradiction:

Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent implements a dynamic anomaly detection system that adapts to different data patterns by using trained neural network models that can handle various time series characteristics including non-seasonal and non-stationary data, unlike static forecast models

Inventive Principle:

Principle #15Dynamics

Solution Approach 2:

The detection system is designed with universal applicability to handle diverse manufacturing processes and sensor data patterns through a unified machine learning framework that works across different data types without requiring model-specific adjustments

Inventive Principle:

Principle #6Universality (Multi-functionality)

4Measurement precision

If supervised regression models are used for multi-dimensional time series data, then prediction capability is improved, but the system complexity increases significantly

Engineering Contradiction:

Improveprediction capabilityVSAvoidsystem complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the complex prediction task into manageable components by training separate neural network models for different aspects of the time series data, allowing accurate multi-dimensional prediction while maintaining system manageability through modular architecture

Inventive Principle:

Principle #1Segmentation

Data Source

PatentUS20250103035A1Method and system for detecting anomaly in time series data

Publication Date: 2025.03.27 ELISA OYJ

AI summary

A method and a system for detecting an anomaly in sensor time series data of a sensing arrangement includes implementing a neural network trained on training data with prior sensor time series data; re-constructing a sample sensor time series data for a target time period using the trained neural network; determining an anomaly score variable based on a re-construction error in the re-constructed sample sensor time series data; determining a confidence interval for the target time period based on a distribution of the determined anomaly score variable; mapping a target sensor time series data, generated by the sensing arrangement corresponding to the target time period, to the determined confidence interval; and indicating an anomaly in the target sensor time series data if the target sensor time series data is not substantially within the determined confidence interval.