Self-Learning Algorithm Calibration for Anomaly Detection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing machine learning approaches for training self-learning algorithms in technical devices face accuracy issues due to reliance on simulation data, which can lead to inaccurate model assumptions, especially when applied to real-world scenarios like detecting anomalies in technical devices.

Innovation Solution

A method that combines basic training with simulation data and build-up training using measured values from the device, allowing the algorithm to be calibrated on a concrete physical instance, thereby improving precision without expanding the database.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of manufacture

If training is performed using only simulation data, then training can be conducted without real-world data collection, but the accuracy and precision of the model predictions deteriorate due to inaccurate model assumptions

Engineering Contradiction:
Improveease of trainingVSAvoidprediction accuracy
Core Design Contradiction:
Ease of manufactureVSMeasurement precision

Solution Approach 1:

The training process is divided into two distinct phases: basic training using simulation data and build-up training using measured data. This segmentation allows each phase to serve its specific purpose - basic training establishes initial model capabilities efficiently, while build-up training refines accuracy with real-world data, thereby resolving the contradiction between training ease and prediction accuracy.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

Basic training using simulation data is performed as a preliminary step before build-up training with measured data. This preliminary action prepares the model with initial knowledge and structures, making the subsequent build-up training more efficient and effective, thus balancing ease of training with final prediction accuracy.

Inventive Principle:
Principle #10Preliminary action

2Measurement precision

If the database is expanded with more real-world data, then model accuracy improves, but data collection time and resources increase

Engineering Contradiction:
Improvemodel accuracyVSAvoiddata collection time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

Instead of requiring extensive real-world data collection, the invention uses a limited set of measured data for build-up training to calibrate the model. This partial action approach achieves sufficient accuracy improvement without the time and resource costs of comprehensive data collection, effectively resolving the contradiction between model accuracy and data collection time.

Inventive Principle:
Principle #16Partial or excessive action

3Productivity

If general model assumptions are used in simulation, then training can proceed with available simulation data, but the applicability to specific real-world instances deteriorates

Engineering Contradiction:
Improvetraining efficiencyVSAvoidapplicability to real instances
Core Design Contradiction:
ProductivityVSAdaptability or versatility

Solution Approach 1:

The build-up training phase adjusts and calibrates model parameters using measured data from specific real-world instances. This parameter change process transforms the general simulation-based model into one that is adapted to specific real-world conditions, thereby resolving the contradiction between training efficiency and applicability to real instances.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentUS20230214660A1Hybrid training method for self-learining algorithms
Publication Date: 2023.07.06 ZF FRIEDRICHSHAFEN AG

AI summary

A method for training a self-learning algorithm, where the algorithm is designed, as a function of one or more physical parameters of a technical device, to predict one or more values of one or more physical parameters of the device. The algorithm undergoes a basic training using values of the physical parameters that have been obtained by simulation of at least part of the device. The algorithm then undergoes build-up training with measured values of the physical parameters.