Why Does a Beautiful DOE Report Fail to Improve Production Yield? — A Case Study on Rebuilding the "Experiment to Implementation" Mechanism in an Electronics Company

1. The 99.7% on the Report, but Not on the Production Line

A wave soldering process in an electronics company has had a long-standing issue with solder bridging, with a defect rate consistently hovering between 1.5% and 2%. The high DPPM (Defects Per Million Opportunities) has led to a continuous stream of customer complaints. Solder bridging is a notorious problem in wave soldering: solder bridges between adjacent pins can cause intermittent functional failures or even entire board scrapping, and the causes are complex—preheating temperature, chain speed, wave height, flux spray volume, and track angle, all of which can contribute to the issue. The company has tried to address this before: engineers increased the chain speed, reducing solder bridging but increasing missed soldering; they increased the flux spray volume, reducing solder bridging but leaving the board dirty with residue. Despite numerous single-factor trials, the defect rate remained unchanged after six months. The process team was then instructed to "solve the problem scientifically," leading to the introduction of DOE (Design of Experiments). Two months later, the team presented a beautiful report: after screening and full factorial experiments on five factors, they found the optimal parameter combination, achieving a pass rate of 99.7% in the trial batch. The management was thrilled and immediately requested a quick transition to mass production. However, the outcome was a cold splash of water—during the first mass production trial, the yield was only 97.2%, more than two percentage points lower than the trial; the next two batches were even worse. The process department claimed the parameters were set by DOE, while the quality department argued that DOE data was "laboratory data" and they couldn't be held responsible. After much debate, the parameters were quietly reverted to their original values, and the DOE report was archived and forgotten. This scenario is common in many companies: the DOE is completed, the conclusions are clear, but the yield remains unchanged. The problem lies not in the experimental design itself, but in the gaps between the experimental conclusions and the implementation in mass production, which no one has managed.

2. The First Hurdle: Experimental Conditions and Production Conditions Are Fundamentally Different

During the post-mortem meeting, the quality manager compared the experimental records with the production records and identified three glaring inconsistencies. Materials: the experiments used newly opened flux and dedicated test boards, while mass production used flux that had been opened for several days and regular batches of boards, which varied in incoming quality. Personnel: the experiments were conducted by process engineers who personally adjusted the machines and monitored the parameters, whereas in mass production, operators followed the usual pace, and parameter drift was not promptly corrected. Equipment: the experiments were conducted right after maintenance, but in mass production, the nozzles had not been maintained for two months. The DOE conclusions were valid under "ideal conditions," but assuming these conditions would hold in mass production is unrealistic. The solution is to incorporate "production achievable conditions" into the factor level table during the design phase—any factors that fluctuate in production, such as material batches, maintenance cycles, and shifts, should either be included in the experiment to assess their impact or at least clearly marked with their sensitivity in the final report. The experimental conditions should align with production conditions, not the other way around. Later, the company added this requirement to the DOE project initiation checklist: during the proposal review, three questions must be answered—what material batches will be used in the experiment? Who will operate the equipment? What is the status of the equipment? Any inconsistencies with the production norm must be justified and assessed in the proposal, and projects that cannot answer these questions are not allowed to proceed.

3. The Second Hurdle: The Optimal Point Often Stands on the Edge of a Cliff

There was another, more subtle issue. During the post-mortem, the engineers noticed that the optimal chain speed in the best combination was at the boundary of the experimental range—higher chain speeds reduce solder bridging, so the optimizer naturally pushed it to the upper limit. However, what lies beyond the boundary is unknown, and even a slight fluctuation within the boundary can drastically worsen the results. This "optimal point" is a peak, not a plateau. A truly robust solution should lie in a relatively flat area of the response surface. The company established a rule: DOE should not only focus on the optimal average value but must also address the stability around the optimal point. Specifically, this involves conducting repeat confirmation tests at the optimal combination and small-scale perturbation tests in the vicinity of the optimal point to observe if the yield drops sharply upon any disturbance. If it does, the team should revert to a suboptimal but more stable combination, and use the signal-to-noise ratio to assist in decision-making if necessary. Every DOE project must include a robustness confirmation record; without this, the report cannot be submitted.

4. The Third Hurdle: No Formal Path from DOE to Change Implementation

Even with reliable conclusions, the implementation must pass through procedural gates: modifying parameters affects the control plan, work instructions, and PFMEA; if the customer is an automotive manufacturer, process changes must be submitted for approval. In the past, engineers would change parameters verbally, taking no credit if successful but bearing the blame if things went wrong, and no one would take responsibility for quality issues—thus, the process department preferred not to make changes. To address this, the company established a standard channel: DOE project conclusion report (including factor effects, robustness confirmation, and risk explanation) → three consecutive pilot batches for verification (using regular material batches, regular operators, and normal production schedules; the pass rate must meet the project target to pass) → change review meeting (signed by quality, process, production, and equipment departments) → ECN (Engineering Change Notice) change → simultaneous updates to the control plan and work instructions → submission to the customer for approval if required → two-month trial operation followed by a review. This channel transformed "parameter changes" from individual actions into organizational processes, ensuring clear responsibility and complete traceability, making it impossible for anyone to "quietly revert" the changes. Each step in the channel has a clear output and a designated signatory: the DOE project conclusion report is prepared by the test engineer, the verification conclusions are reviewed by the quality engineer, the change review is led by the process manager, and the updates to the control plan and work instructions are verified by the document administrator—any issues can be traced back to the responsible person, and the credit for successful changes is clearly distributed.

5. After Rebuilding: From 1.8% to 0.2%, from a Single Case to a Routine

Following these three steps, the team re-ran the process: the experimental design was changed to "screening + robust window selection," the chain speed was reduced by about 10% from the boundary, and the flux spray volume and preheating temperature were set in a flat region. The three pilot batches passed the verification with a stable pass rate of 99.5%. After completing the ECN, the mass production solder bridging defect rate dropped from 1.8% to around 0.2%, remaining stable for three consecutive months, and the customer DPPM (Defects Per Million Opportunities) significantly decreased. More valuable than the numbers is the mechanism that was established: over the next year, the company replicated this path in six other processes, including reflow soldering, dispensing, and screw locking. DOE transformed from a "one-time effort" into a "continuous improvement" tool, and the cycle from experiment to mass production release was reduced to an average of four weeks. The quality manager noted during the post-mortem: "We never lacked DOE skills; what we lacked was a channel to implement DOE conclusions on the production line." He also observed a subtle change: previously, the equipment department considered DOE the responsibility of the process department, but now they actively align maintenance schedules with the experimental windows, as everyone has experienced the benefits of "getting it right the first time and reducing rework"; front-line team leaders are also willing to check the new parameter cards during morning meetings, as they have personally participated in the pilot verification and understand the origin of these numbers.

6. Four Gates to Ensure DOE Conclusions Enter the Production Line

The lessons learned from this company can be distilled into four gates that any DOE project should pass before conclusion. Gate 1, Condition Consistency: Are the materials, equipment, personnel, and environment in the experiment the same as those in mass production? Have inconsistent factors been included in the experiment or clearly marked with their sensitivity? Gate 2, Robustness: Is the optimal point standing on the edge of a cliff? Have neighboring perturbation and repeat confirmation tests been conducted? Gate 3, Pilot Verification: Have three consecutive pilot batches passed verification under regular conditions? Has the yield met the project target? Gate 4, Change Closure: Have the parameters been processed through ECN? Have the control plan, work instructions, and PFMEA been updated? Have the necessary changes been submitted to the customer for approval? Only when all four gates are passed can the DOE be considered truly "complete"; if any gate is missing, no matter how beautiful the report, it remains just a theoretical conclusion.

7. One Sentence Summary

The endpoint of DOE is not a beautiful analysis report, but a stable yield on the production line—what often falls short is not statistical techniques, but the three bridges of condition consistency, robustness confirmation, and change closure.

The value of DOE lies not in how beautiful the report is, but in whether the conclusions can be stably implemented on the production line—before implementation, pass through the four gates.

Knowledge code: 6.4.1

Version: v20260905

Author: QTank QTank is dedicated to providing systematic professional knowledge, methodologies, and practical tools for quality management practitioners, helping companies continuously improve their quality capabilities.