Machine Safeguarding via Sensor Quality Scoring

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing safety technologies using neural networks for machine monitoring struggle to reliably assess the quality of sensor data, leading to potential misjudgments and unsafe conditions due to poor-quality images, especially in dynamic scenarios, without providing sufficient training data for anomaly detection.

Innovation Solution

A method using supervised machine learning, specifically neural networks, to evaluate sensor data quality by training with perturbed examples to determine a quality score, ensuring reliable detection capability testing and safeguarding against unsafe conditions.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Difficulty of detecting and measuring

If neural networks are used to evaluate sensor data for safety monitoring, then detection capability is improved, but reliability deteriorates due to inability to assess data quality and potential false negatives

Engineering Contradiction:
Improvedetection capabilityVSAvoidsafety assessment reliability
Core Design Contradiction:
Difficulty of detecting and measuringVSReliability

Solution Approach 1:

The patent implements a quality assessment module that provides feedback on sensor data quality before neural network evaluation. This feedback mechanism analyzes whether sensor data meets minimum quality thresholds (completeness, clarity, absence of artifacts) and prevents unsafe evaluations from proceeding, thereby maintaining reliability while preserving detection capability.

Inventive Principle:
Principle #23Feedback

Solution Approach 2:

The patent performs preliminary quality assessment of sensor data before it is fed to the neural network for safety-critical evaluation. By pre-screening data for adequacy, completeness, and quality thresholds, the system ensures that only suitable data undergoes neural network analysis, preventing false negatives from poor-quality input.

Inventive Principle:
Principle #10Preliminary action

2Reliability

If conventional safety sensors are used with redundant electronics, then reliability is improved, but device complexity increases

Engineering Contradiction:
Improvesensor reliabilityVSAvoidelectronic system complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent replaces complex redundant electronic safety systems with a software-based quality assessment module that runs on standard computing hardware. This module evaluates sensor data quality through algorithms analyzing completeness, clarity, and artifact detection, achieving reliable safety assessment without additional redundant electronics or complex hardware architectures.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

3Reliability

If protective fields are configured to prevent intrusion, then safety is improved, but productivity deteriorates due to system stops

Engineering Contradiction:
Improvesafety protectionVSAvoidmachine availability
Core Design Contradiction:
ReliabilityVSProductivity

Solution Approach 1:

The patent applies partial quality assessment by evaluating only critical quality thresholds that are sufficient for safe neural network operation, rather than implementing comprehensive analysis of all possible sensor data aspects. This selective approach ensures safety requirements are met while minimizing processing overhead and avoiding unnecessary system stops, thereby maintaining productivity.

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP4553366B1Securing of a machine
Publication Date: 2025.10.08 SICK AG
  • EP4553366B1 patent drawingFigure 1~2
  • EP4553366B1 patent drawingFigure 3

AI summary

A method for safeguarding a machine (12) is described, in which a sensor (10) monitors the machine (12) and generates sensor data that is evaluated so that a hazardous situation is detected and, in the event of a hazardous situation, the machine (12) is safeguarded. A detection capability test verifies whether an assessment of a hazardous situation is possible; otherwise, the machine (12) is safeguarded. In the detection capability test, the sensor data is evaluated using a machine learning method with at least one quality score, and an assessment of a hazardous situation is only considered possible if the quality score is sufficient.