Kernel-Based Job Scheduling for Adaptive Manufacturing Priorities
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing scheduling systems for manufacturing and logistics processes face challenges in determining efficient priority functions that adapt to specific constraints and changes over time, leading to suboptimal job scheduling.
Innovation Solution
The use of kernel-induced priority functions that compare jobs to reference jobs, allowing for the definition of expressive and adaptable priority rules through machine learning techniques, enabling efficient scheduling by computing similarity and optimizing priority values based on training data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional list scheduling with fixed priority functions is used, then computational efficiency is improved, but scheduling performance and adaptability to specific process constraints deteriorate
Solution Approach 1:
The patent applies dynamics by transitioning from fixed, static priority functions to dynamic, learned priority functions that adapt to changing process constraints. The system continuously learns optimal scheduling policies from process data, enabling the scheduling mechanism to dynamically adjust to specific manufacturing or logistics constraints while maintaining computational efficiency through the use of learned models.
Solution Approach 2:
The patent changes the parameters of the priority function from predetermined fixed values to learned parameters that are optimized based on process performance data. By using machine learning techniques to determine priority function parameters, the system achieves both computational efficiency and adaptability to specific process constraints, resolving the contradiction between these two requirements.
2Productivity
If custom priority functions tailored to specific processes are developed, then scheduling performance is improved, but computational complexity and determination time worsen
Solution Approach 1:
The patent applies preliminary action by pre-learning optimal priority functions offline using historical process data before actual scheduling operations. This offline training phase captures process-specific constraints and optimization criteria, so that during runtime, the system can directly apply the pre-learned priority functions without complex real-time computations, thus achieving high scheduling performance with reduced computational complexity.
Solution Approach 2:
The patent uses copying by creating simplified models of complex scheduling problems through learned priority functions. Instead of solving complex optimization problems in real-time, the system learns from examples and copies the essential scheduling patterns into compact priority functions that can be evaluated efficiently, reducing computational complexity while maintaining scheduling performance.
3Adaptability or versatility
If priority functions are frequently recomputed to adapt to changing constraints, then adaptability is improved, but computational overhead and processing time worsen
Solution Approach 1:
The patent applies feedback by implementing a learning mechanism that continuously monitors process performance and constraint changes, then updates priority functions accordingly. The system uses feedback from scheduling outcomes and changing process conditions to adaptively refine priority functions, achieving high adaptability without requiring frequent complete recomputations, thus reducing processing time while maintaining responsiveness to constraint changes.
Data Source
AI summary
A computer-implemented method of scheduling jobs of a manufacturing or logistics process using a priority function. The priority function is evaluated on multiple to be scheduled jobs to obtain multiple respective priority values. The priority function is defined to invoke a kernel function. The kernel function is defined to compare representations of two respective jobs. Evaluating the priority function on a selected job comprises evaluating the kernel function on representations of the selected job and one or more reference jobs. The schedule for the multiple jobs is determined based on the priority values and is then output to enable the multiple jobs to be carried out according to the schedule.