At TT PSC, we believe that manufacturing leaders should rely on reliable, real-time data to achieve the real goals of Industry 4.0 (r)Evolution. This data can build knowledge and awareness and identify actions requiring improvement.
In the Lesson Learned Explained series, we present examples of completed projects along with insights. We hope that this short material will help you avoid at least some of the pitfalls on the way to Digital Transformation.
Challenge: Monitor and improve the quality of end products
A company in the electronics industry specializing in semiconductor manufacturing set a major goal to make improvements that positively affect the quality of final products. A key element was to monitor and identify correlations that would predict the satisfactory quality of products coming off the production line. This was done using data from machines and quality control stations, which was then subjected to in-depth analysis. This enabled the company to better understand which factors affect the quality of their products.
Solution: Systems integration and data modeling
To achieve its goal, the company, in cooperation with TT PSC, implemented a solution that included integration with PLCs and production support systems. A key element was performing historical data analysis and modeling with predictive quality capabilities. This made it possible to predict potential quality problems, catch them quickly, solve them, and prevent their occurrence in the future.
Implementation Problems: Low accuracy in AI modeling.
Despite advanced AI (Artificial Intelligence) modeling based on process parameters, the solution proved ineffective and had low accuracy. The problem was identified as too many factors—monitored and unmonitored—and various product variants, making more complex models costly and inefficient. As a result, the algorithms could not accurately predict the quality of the products after the manufacturing process.
Correction: A new model based on physical equations
In response to the identified problems, TT PSC experts built a new model based on physical equations. Instead of relying solely on artificial intelligence (AI) and machine learning (ML) algorithms to analyze process data, they used an approach grounded in fundamental physics principles.
How it could positively impact the problem:
- Complexity reduction: Models based on physical equations can be less complex than advanced AI models because they are based on well-understood physics principles rather than analyzing large amounts of data. This reduces the risk of errors due to an excessive number of variables.
- More straightforward interpretation: Physical models are often more transparent and easier to interpret than AI models, allowing engineers to better understand what factors affect product quality.
- Stability and consistency: Physical models can be more stable and consistent because they are based on fixed laws of physics rather than on variable process data that can vary from one product series to another.
- Scalability: Physical models can be more easily scaled and adaptable to different product variants, which is more difficult to achieve with complex AI models.
This enabled more accurate product quality predictions, eliminating the influence of excessive variables and product series differences. This allowed for more consistent and reliable results.
Result: Successful identification and resolution of quality challenges
Initially, the ML model used to analyze process parameters could not achieve the expected accuracy. This was due to too many factors and differences between product series. Only by using standard physics-based modeling equations could the AI/ML algorithm work effectively. As a result, the company could control the quality of its products better and react faster to potential problems.
Conclusions
Several key conclusions can be drawn from the activities carried out:
- Even advanced algorithms should generate results that are easy to interpret and practical.
- Algorithms should support the decision-making process by providing valuable information.
- Industry expertise and the experience of manufacturing technologists are key to successfully implementing machine learning (ML).
- Algorithms can analyze data that experts’ intuitive approach would typically overlook.
_Are you interested in digital transformation for your enterprise?
Check out other implementation examples below and the lessons learned from them.
