.png){kind=icon}
.png){kind=icon}
Return
An evolutionary multi-objective ensemble learning algorithm for quality prediction in complex manufacturing processes
.png){kind=icon}
PRE
AI
DOI：10.1080/00207543.2026.2718791.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
In modern industry, quality prediction (QP) in complex manufacturing processes (CMPs) is essential for controlling the quality of complex products because direct quality measurement is often time-consuming and costly. However, data-driven QP methods face challenges in handling CMP data with high-dimensional process variables, complex intervariable interactions, and a limited number of labelled samples. To address these issues, this paper proposes EMEL-NB, a novel evolutionary multi-objective ensemble learning algorithm for QP with a two-phase model construction procedure. First, EMEL-NB integrates a boosting-inspired sample-weight adjustment strategy with an evolutionary feature selection algorithm to generate genotypes for sample-efficient naive Bayes base learners, promoting learner diversity across both feature and sample spaces. Second, a sparse ensemble is constructed using an <span class="NLM_disp-formula-image inline-formula rs_preserve"><img src="//:0" alt="" class="mml-formula" data-formula-source="{"type":"image","src":"/cms/asset/adb236f7-eb5b-49a5-93f1-202ae3161446/tprs_a_2718791_ilm0001.gif"}"></span><span class="NLM_disp-formula inline-formula rs_preserve"><img src="//:0" alt="" data-formula-source="{"type":"mathjax"}">
<math>
<msub>
<mi>L</mi>
<mn>1</mn>
</msub>
</math></span>-regularised logistic regression meta-learner. A knowledge transfer strategy is proposed to train the meta-learner on evaluation data during the FS process to improve its generalisation performance. Experimental results on real-world CMP datasets show that EMEL-NB achieves better overall predictive performance than the other methods evaluated in this study. It also maintains low computational cost during online prediction, supporting its potential use in industrial QP applications. Furthermore, the proposed algorithm supports model interpretation by quantifying the importance of process variables.
Keywords:
Quality prediction
ensemble learning
multi-objective evolutionary algorithms
feature selection
complex manufacturing processes
Quality control
evolutionary computation
genetic algorithms
Journal
IF:
7.3
Papers:
1.1W
Citations:
3.7W