Robot Process Quality Assurance Using Machine-Learned Virtual Sensing

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing robot-supported processes face challenges in quality assurance, particularly in reliability, speed, and range of application, due to limitations in measuring and improving process quality effectively.

Innovation Solution

A method utilizing a machine-learned process model to classify and adjust process success, incorporating robot-specific data from sensors, enables improved quality assurance by classifying processes into discrete or continuous success categories, allowing for real-time adjustments and reducing manual data selection, thereby enhancing reliability and flexibility.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If traditional quality assurance methods are used for robot-assisted processes, then the system is simpler to implement, but the reliability and accuracy of quality assessment deteriorates

Engineering Contradiction:
Improvequality assurance reliabilityVSAvoidsystem complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent replaces traditional mechanical measurement systems with a machine learning-based virtual sensor model. The system uses data from existing robot sensors (encoders, current sensors) to train a neural network that virtually measures process quality without physical measurement devices. This substitution maintains reliability while reducing device complexity.

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

Solution Approach 2:

The patent creates a virtual copy of the measurement function through machine learning. Instead of physical sensors directly measuring quality parameters, the system trains a model that copies the measurement capability by learning from sensor data and process outcomes. This virtual copying enables accurate quality assessment without additional physical measurement hardware.

Inventive Principle:
Principle #26Copying

2Productivity

If manual selection of process data is used, then the system is easier to configure, but the speed and efficiency of quality assurance deteriorates

Engineering Contradiction:
Improvequality assurance speedVSAvoidmanual data selection effort
Core Design Contradiction:
ProductivityVSEase of operation

Solution Approach 1:

The system performs self-configuration through automated machine learning model training. The neural network automatically identifies relevant process parameters and selects optimal data sources during the training phase, eliminating the need for manual data selection. The system serves itself by learning which sensor data correlates with quality outcomes, thereby increasing speed without sacrificing ease of operation.

Inventive Principle:
Principle #25Self-service

3Measurement precision

If additional sensors are deployed for quality measurement, then measurement precision improves, but device complexity and cost increase

Engineering Contradiction:
Improveprocess quality measurement precisionVSAvoidsensor system complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent replaces physical measurement sensors with a virtual sensor implemented as a machine learning model. The neural network processes data from existing robot control sensors to predict quality parameters, achieving measurement precision without adding physical measurement devices. This substitution eliminates the need for additional sensors while maintaining or improving measurement accuracy.

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

Solution Approach 2:

The system makes existing robot sensors serve multiple functions: both robot control and quality measurement. By training the machine learning model to interpret data from standard robot sensors (encoders, current sensors), the system enables these sensors to provide both motion control information and quality assessment information, eliminating the need for dedicated measurement sensors.

Inventive Principle:
Principle #6Universality (Multi-functionality)

4Adaptability or versatility

If extensive process data is collected and processed, then the scope of application improves, but computing time increases

Engineering Contradiction:
Improverange of applicationVSAvoidcomputing time
Core Design Contradiction:
Adaptability or versatilityVSLoss of time

Solution Approach 1:

The system performs preliminary data processing during the model training phase. The neural network is trained offline on extensive historical process data, learning patterns and relationships in advance. During actual quality assurance operations, the pre-trained model quickly processes new data without requiring extensive real-time computation, thereby expanding application scope while minimizing computing time delays.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP3710207B1Quality assurance of a process
Publication Date: 2023.08.23 KUKA DEUT GMBH

AI summary

The invention relates to a method for providing quality assurance of a process comprising the following steps repeated a plurality of times: - executing (S10) the process; - detecting (S10) learning process data during this execution of the process; and - detecting (S20) a learning evaluation of the executed process; the step: - machine-learning (S30) a model (10) of the process on the basis of the detected learning process data and learning evaluations; and the steps, in particular repeated a plurality of times: - renewed execution (S50) of the process; and - detecting (S50) current process data during this renewed execution of the process; wherein, on the basis of the current process data and the machine-learned model (10) - a success of the newly executed process is determined, in particular classified (S60), and/or - the newly executed process is modified (S60).