Management Decisions in Manufacturing using Causal Machine Learning -- To Rework, or not to Rework?

TL;DR

This work addresses how to decide whether to apply inline rework in a multistage, lot-based manufacturing setting, where final yield $Y$ is only known after inspection. It develops a causal framework based on double/debiased machine learning (DML) to estimate conditional average treatment effects (CATE) and learns policies $\pi(X)$ that balance rework costs with yield gains, accounting for confounding via observed lot and system states. The authors validate the approach on a real AMS-Osram LED phosphor-conversion dataset, showing that data-driven policies can achieve yield improvements of about $2$–$3\%$ over naive strategies, with robustness checks indicating insensitivity to reasonable unobserved confounding. The contribution combines a rigorous causal identification strategy with practical policy-learning methods, enabling per-lot, state-aware rework decisions that improve production efficiency in complex, multi-stage manufacturing.

Abstract

In this paper, we present a data-driven model for estimating optimal rework policies in manufacturing systems. We consider a single production stage within a multistage, lot-based system that allows for optional rework steps. While the rework decision depends on an intermediate state of the lot and system, the final product inspection, and thus the assessment of the actual yield, is delayed until production is complete. Repair steps are applied uniformly to the lot, potentially improving some of the individual items while degrading others. The challenge is thus to balance potential yield improvement with the rework costs incurred. Given the inherently causal nature of this decision problem, we propose a causal model to estimate yield improvement. We apply methods from causal machine learning, in particular double/debiased machine learning (DML) techniques, to estimate conditional treatment effects from data and derive policies for rework decisions. We validate our decision model using real-world data from opto-electronic semiconductor manufacturing, achieving a yield improvement of 2 - 3% during the color-conversion process of white light-emitting diodes (LEDs).

Figures (11)

• Figure 1: Exemplary linear setup with manufacturing stages $S_1, \ldots, S_n$ and a final inspection stage $\mathop{\mathrm{FI}}\limits$. Decisions at $S_k$ may depend on previously observed product state $X_p$ and the overall system state $X_s$.
• Figure 2: Manufacturing $M$, inline inspection $\mathop{\mathrm{I}}\limits$ and decision step $\mathop{\mathrm{OR}}\limits$ at stage $S_k$.
• Figure 3: Graphical Causal Model: Yield $Y$ and rework decision $A$ are confounded by lot state $P$ and system sate $S$.
• Figure 4: CIE 1931 color space chromaticity diagram with exemplary color points $C_0$ of the ground substrate, $C_1$ after conversion, and $C_2$ for the final product.
• Figure 5: Histogram of observations in the "treatment" and "no treatment" groups of different covariates. In general, the overlap assumption appears to hold.