Automation Context Modeling for Concept Drift in Time Prediction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Manufacturing systems face challenges in accurately estimating processing times due to high volatility and dynamics in markets, with existing analytical approaches failing to account for situational dependencies like changeovers and material quality, leading to inefficient production plans and schedules.

Innovation Solution

A context-aware analytics framework that integrates context knowledge and history into manufacturing operations management systems, using a learning-based prediction model to detect concept drifts and adapt analytic models dynamically, ensuring accurate and reliable processing time estimates by extending the current context with adjacent data from a context knowledge base.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If fixed or stationary distributed processing times are used in analytical approaches or simulation models, then the production planning and scheduling can be implemented, but the estimates become inaccurate when situational dependencies (changeovers, maintenance events, material quality) significantly influence machine processing times

Engineering Contradiction:
Improveprocessing time estimation accuracyVSAvoidability to handle situational dependencies
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent transforms static processing time models into dynamic ones by continuously adapting the prediction model to current manufacturing situations. The system monitors contextual parameters (changeovers, maintenance events, material quality) and dynamically adjusts processing time estimates, allowing the model to respond to situational dependencies rather than relying on fixed historical averages.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent changes the parameters used in processing time estimation from fixed historical values to dynamic parameters that incorporate contextual information. By introducing contextual parameters (product type, material quality, machine state) into the estimation model, the system adapts processing time predictions to current manufacturing conditions, resolving the contradiction between accuracy and adaptability.

Inventive Principle:
Principle #35Parameter changes

2Loss of information

If manual monitoring of process variables and manual specification of events by engineers is performed, then context-aware analysis can be achieved, but the approach becomes inefficient in frequently changing environments with varying orders, suppliers, and devices

Engineering Contradiction:
Improvecontext information captureVSAvoidreconfiguration efficiency
Core Design Contradiction:
Loss of informationVSProductivity

Solution Approach 1:

The patent implements self-service by enabling the system to automatically acquire, process, and learn from contextual information without manual intervention. The prediction model automatically adapts to changing environments by learning from observed patterns in processing times and contextual parameters, eliminating the need for engineers to manually reconfigure the system when orders, suppliers, or devices change.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The patent incorporates feedback mechanisms where the system continuously monitors actual processing times and contextual parameters, compares predicted vs. actual values, and uses this feedback to refine and adapt the prediction model. This closed-loop approach enables automatic adaptation to changing environments, maintaining context awareness without manual reconfiguration efforts.

Inventive Principle:
Principle #23Feedback

3Adaptability or versatility

If offline analysis with pre-defined simulation models is performed, then adaptive learning can be implemented, but the models fail to account for concept drifts in non-stationary distributions of machine processing times

Engineering Contradiction:
Improveadaptive learning capabilityVSAvoidmodel accuracy under concept drift
Core Design Contradiction:
Adaptability or versatilityVSReliability

Solution Approach 1:

The patent transitions from static offline simulation models to dynamic online prediction models that continuously adapt to concept drifts. The system processes data in real-time and updates the prediction model as new contextual patterns emerge, enabling the model to maintain accuracy even when the underlying data distribution changes due to concept drift in processing times.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent prepares the system for concept drifts by pre-configuring the prediction model with contextual parameters and learning mechanisms, but the actual adaptation occurs dynamically when drifts are detected. The system maintains a ready-to-adapt state by continuously monitoring contextual parameters, allowing it to respond quickly to concept drifts without requiring complete model redefinition.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentEP3256993B1Method and apparatus for operating an automation system
Publication Date: 2021.06.16 SIEMENS AG
  • EP3256993B1 patent drawingFigure 1
  • EP3256993B1 patent drawingFigure 2
  • EP3256993B1 patent drawingFigure 3~4

AI summary

The present invention relates generally to a method and an apparatus for operating an automation system. The method for operating an automation system comprises the method steps of: a) providing a learning-based prediction model (M) for the automation system trained by process data (D) comprising context of an automation process, b) receiving information about current context of the automation process, c) verifying context change by comparing the current context to the context of said process data, d) in the case of any context change verifying a concept drift by comparing pre-drift process data and post-drift process data, e) in the case of any concept drift re-training said model (M) with post-drift process data, f) in the case of no context change testing for random concept drift not detected by verifying context change, g) in the case of any random concept drift extend the current context by using data comprising previous context changes, h) otherwise no further method steps are required.