Issue link: https://iconnect007.uberflip.com/i/1539960
66 SMT007 MAGAZINE I OCTOBER 2025 fying these issues is time-consuming, especially when there is a large amount of equipment. An AI degradation detection algorithm was developed to learn from the measurement values and highlight the deterioration trend, avoiding unex- pected failures and downtime. The model is based on an unsupervised decision tree learning algorithm. When the model was implemented in actual pro- duction, its predictive capability enabled test engi- neers to receive real-time alerts to undertake pre- emptive maintenance before tests started to fail. This lowered the retest rates from 7% down to just 1%, along with significant savings in debugging efforts. Based on the insights, the engineering team was also able to optimize their probe maintenance process and change only the necessary probes. This further reduced probe spending by 20%, trans- lating into $3,000 yearly savings per machine. Use Case#3: Longer NPI lifecycle Due to rapid technological advancements cou- pled with frequent new product releases from busi- nesses, manufacturers are pushed to continuously introduce new product introduction (NPIs) in their production lines for mass production. Taking new NPIs into production involves working with multiple vendors, resulting in longer lead times and limiting the number of NPIs that can be rolled out in a year. Any improvement to reduce lead times can help launch new products to market in a shorter time frame, hence positively contributing to the top-line revenue. We implemented an AI/ML-based solution that helped customers fast-track the NPI cycle and reduce the long lead time from designing the fixture to taking it for mass production. NPI rollouts took around three months before getting into mass production. One of the primary reasons was the fine-tuning of tests to achieve a minimum process capability index (Cpk) value of 1.33. Cpk scores are one of the aspects of man- ufacturing that showcase the process's stability. The higher the Cpk scores, the better the yield quality produced. Fine-tuning the Cpk for new fix- tures involved collaborating with the fixture house to identify and address tests with low Cpk val- ues iteratively. This process required redesigning fixtures and programs. Multiple builds are com- pleted before a product reaches the mass produc- tion stage. During each build, in-circuit test (ICT) measurements are manually extracted, computed, and analyzed to improve the subsequent build's first-pass yield (FPY). These manual processes P u r p l e d ot s a re d e g ra d i n g a n o m a l i e s p re d i cte d by A I m o d e l b efo re t h e a ct u a l fa i l u re s . ▼