Gu Niansong | AI+DFSS Design for Six Sigma: From Engineering Theory to Cloud Implementation

On May 9, the "Light Journey Together Resource Gathering Series" salon, jointly created by China Light Media · Ming Classroom and Ningbo Lighting Exhibition, was successfully held. The event specially invited Mr. Gu Niansong, a senior expert in R&D and innovation, to focus on the theme of DFSS Design for Six Sigma — AI Empowerment: From Engineering Theory to Cloud Implementation. Combining nearly 30 years of cross-industry R&D practical experience, he dissected the core logic of DFSS, tolerance simulation practices, AI-DFMEA, and VOC intelligent insights, helping lighting companies bridge the full chain from "engineering theory → tool implementation → cloud empowerment", providing a complete implementation solution to control risks, improve efficiency, and align with the market from the design source.

Re-understanding:

Break cognitive misconceptions, return to essential value

At the opening of the event, Teacher Gu first clarified common misconceptions in the industry regarding DFSS (Design for Six Sigma):

Many people believe that DFSS is about pursuing a defect rate of 3.4 ppm, but this is actually a "myth"—from a statistical perspective, considering a long-term drift of 1.5σ, 3.4 ppm corresponds to a one-sided 4.5σ level, with a Cpk of approximately 1.5, rather than the theoretical .What truly matters is not the absolute ppm value, but the stability of the design under actual manufacturing variations.

The core theoretical foundation of DFSS is the Central Limit Theorem: products are assembled from multiple components, each with its own tolerances and variations. Regardless of the distribution shape of individual parts, as long as there are enough of them, the overall variation after assembly will necessarily follow a normal distribution.

The core issues that DFSS aims to address are: how to achieve robust product design given manufacturing variations in components, how to develop the right products that truly meet market needs, and how to control R&D risks from the source.

To date, DFSS has evolved into a complete product development methodology framework, namely the DMADV process:

• Define: Clarify product positioning and market demand to avoid the tragedy of "great technology but poor sales"

• Measure: Quantify customer requirements into actionable product specifications and key quality characteristics ()

• Analyze: Identify design risks and key influencing factors using tools such as DFMEA

• Design: Optimize solutions using robust design, tolerance analysis, Monte Carlo simulation, and other methods

• Verify: Ensure the design meets requirements through reliability testing, CPK verification, etc.

Teacher Gu particularly emphasized: "DFSS is not a creativity engine, but an engine for risk mitigation and rework reduction; it is not exclusive to statisticians, but a practical method for every engineer to understand the 'Y = f(x)' causal relationship; it also does not require absolute 6σ capability, but helps enterprises find the optimal balance between cost and robustness."

Implementation of Core Methodology

From Tolerance Analysis to Risk Control, Solving R&D Pain Points

Four Types of Tolerance Analysis Methods,Solving Pain Points in Complex Scenarios

Tolerance accumulation after product assembly is the core cause of low mass production yield and reliability issues. Teacher Gu compared four common tolerance analysis methods:

1. Worst-case method: Assumes all tolerances are simultaneously at their limits; safest but overly conservative, with the highest cost

2. Root Sum Square (RSS) method: Based on the normal distribution assumption, calculates the square root of the sum of squared tolerances; more suitable for most mechanical assembly scenarios

3. Average of Worst-case and RSS methods: A compromise between the two methods, balancing safety and rationality

4. Monte Carlo simulation: Supports arbitrary distributions and nonlinear relationships through random sampling of parameters; the preferred solution for complex systems (such as optical alignment and snap-fit assembly)

Teacher Gu shared a practical case: A water trigger leakage issue in a floor scrubber, where about 1 out of 100 units leaked, with the root cause remaining elusive for a long time. After 5,000 Monte Carlo simulation trials, it was found that two parameters contributed 85% of the gap variation. Tightening the tolerances of these two key components completely solved the problem, avoiding the high costs of repeated mold modifications.

"In the past, similar issues took 6 months and three mold revisions to resolve. With Monte Carlo simulation, the root cause can be identified in just 2 days, delivering a very high ROI."

AI-powered DFMEA addresses traditional pain points to achieve proactive risk management

DFMEA (Design Failure Mode and Effects Analysis) is a core tool for R&D risk management, but traditional DFMEA has three major pain points:

• Heavy reliance on individual engineer experience, making it difficult for junior engineers to cover potential risks

• Lack of an internal historical failure database, leading to loss of experience when personnel leave

• A complete DFMEA requires a week or even longer, often becoming a document filled in after the fact

The AI-DQA (Design Quality Assurance) tool developed by Teacher Gu's team perfectly solves these problems by achieving automated risk identification through a four-step process:

1. Automatic product decomposition: Automatically breaks down the input product name and description into a three-level structure of "Function → Module → Part"

2. Multi-source knowledge fusion: Integrates built-in product templates, enterprise internal failure databases, industry knowledge graphs, and real-time failure cases from across the web

3. AI intelligent analysis: Calculates Risk Priority Number (RPN) based on DFMEA principles and automatically ranks the Top 5 high-risk items

4. Output Mitigation Solutions: Provide specific prevention and improvement measures for each high-risk point

Practical Case: In a smart home hub project, traditional DFMEA overlooked the heat dissipation risk of the power regulator. The AI tool identified it as the second-highest risk by mining similar product failure databases. Adding just a simple thermal pad avoided potential recall losses of $2 million, delivering a return on investment hundreds of times the tool's value in this single case.

AI Captures Voice of Customer (VOC) to Build the "Right Product" from the Source

"The biggest pain point in my 30 years of R&D is that many products have excellent technical specifications, but the market simply doesn't buy them." Teacher Gu candidly stated that traditional VOC methods (surveys, focus groups) are not only costly and time-consuming but also prone to bias, leading to a disconnect between product definition and actual market needs.

AI-drivenProduct Feasibility Analysis Tool enables efficient mining of user needs: By simply entering basic information such as product name, description, target market, and sales channel, the system automatically captures massive user reviews and social data from platforms like Amazon and Xiaohongshu, outputting a complete market analysis report within minutes, including:

• Market Trends: Determine whether the industry is in a growth, stable, or decline phase

• User Pain Points: Extract high-frequency dissatisfaction from real reviews and convert them into actionable technical specifications

• Competitor Analysis: Outline the strengths, weaknesses, and price ranges of mainstream competitors

• Market Gaps: Identify differentiated opportunities that users need but are not yet met

• Risk Assessment: Quantify technical and market risks and provide entry recommendations

"I once led a floor scrubber project that achieved 6σ reliability, but it ultimately failed commercially because we missed the user trend towards wireless models. If we had this AI tool back then, we could have identified this trend from market data in just one hour, avoiding huge investment losses." Teacher Gu emphasized, "Even the best DFSS execution cannot save a product nobody wants. Do not start any design before quantifying market demand."

Cloud-based SaaS Toolset

Lower the barrier to entry and enable end-to-end engineering empowerment

Teacher Gu’s team has packaged 30 years of engineering experience into an out-of-the-box AI-DFSS SaaS toolset, including four core tools: tolerance simulation, AI-DFMEA, AI-VOC analysis, and failure root cause analysis. This reduces analysis work that originally took days to just minutes, significantly lowering the threshold for using statistical methods, allowing ordinary engineers to get started quickly and empowering engineering decision-making across the entire organization.

Core Advantages of the Tools:

✅ Full-process coverage: From market demand insights and design risk prevention to root cause analysis of mass production issues, forming a complete R&D closed loop

✅ Multilingual support: The tool interface and reports support switching between Chinese and English, meeting the needs of global enterprises

✅ Globally available: Servers are deployed overseas, allowing users both domestically and internationally to access and use directly without additional network configuration

✅ Continuous iteration: The toolset is continuously updated, currently adding industry-specific tools such as Healthy lighting metric calculation

Interactive Q&A

Do I need to queue to use the tool?

No, simply enter the parameters to run the simulation and get instant results.

Does the tool support multilingual reports?

Currently, Chinese and English are supported. The report is generated in the language selected on the portal. Other languages can be exported and translated manually.

Is special network configuration required for domestic access?

Not required. The server is deployed overseas and can be accessed directly from both domestic and international locations. Domestic usage may experience slightly slower loading speeds, but this does not affect normal functionality.

What other application directions does AI have in the lighting industry?

Current LED luminous efficacy is already approaching the physical limits of mass production. The future growth direction of the lighting industry mainly lies in light quality and intelligence. AI can help enterprises optimize light recipe design, intelligent control algorithms, and healthy lighting parameter matching, making it the core competitiveness for the next stage of the industry.

Conclusion

Embrace AI + DFSS to improve R&D quality and efficiency

In his summary, Teacher Gu emphasized: "AI is a true industrial revolution that can increase productivity a hundredfold; failing to embrace it will inevitably lead to elimination. DFSS is not an lofty theory, nor is AI an unreachable concept. The toolset combining both can practically help enterprises solve R&D pain points, achieving full-chain optimization from 'making the right product' to 'making the product right'."