Issue link: https://iconnect007.uberflip.com/i/1539960
OCTOBER 2025 I SMT007 MAGAZINE 69 A I I m p l e m e nt at i o n a rc h i te ct u re fo r a t y p i c a l S M T l i n e. ▼ ture with high processing capacity and substantial data storage for retaining historical data. If the organization is open to sharing data out- side their environment, they can opt for analytics- as-a-service platforms that handle and manage all the data processing. This can significantly save hardware costs and reduce efforts in infrastructure setup. Most of the organizations we engage have limited IT resources and opt for the analytics-as- a-service option due to these benefits, along with the end-to-end professional guidance (such as user training, process optimizations, etc.) they receive for rolling out such initiatives successfully. The most common implementation architecture for AI solutions involves leveraging the existing dig- ital infrastructure used by the MES system to con- nect to equipment. The same network can be used for transferring the log files generated by machines to the analytics-as-a-service application. Analytics results are consumed by users via a separate UI (user interface) or can be integrated with MES via APIs (application programming interface). While it is recommended to keep MES separate from the Ana- lytics platform, as an Analytics platform is continu- ously evolving and might impact the performance of MES. Challenges for Implementing AI Deploying AI/ML models in a practical manufac- turing setting indeed comes with its own set of challenges. Implementing a successful AI/ML initiative as a fully operational production tool requires extensive planning and foresight. Here are some key challenges to consider before starting an AI initiative: 1. High cost for in-house implementation: Organizations planning to implement AI/ ML solutions in-house will first need to set up complete hardware infrastructure. This requires significant capital expenditure and involvement from multiple teams, including IT Infrastructure, IT Security, and Procurement. Additionally, a software team with multiple skill sets needs to be hired and managed. 2. Adoption and data literacy: One of the most common reasons for unsuccessful AI/ML ini- tiatives is the low adoption of AI/ML results. This can be attributed to a lack of data liter- acy. Users often struggle to understand or interpret the results from analytical models and hence avoid using them. Special focus should be given to a rollout plan that includes creating "data-savvy" users and ensuring peri- odic checkpoints to measure success. 3. Advancements in use cases: There are rapid advancements in AI/ML models in terms of their accuracy, sophistication, and perfor- mance. Any model currently running in pro- duction needs to be re-trained or revisited periodically to ensure model accuracy. Orga- nizations implement MLOps (Machine Learn- ing Operations) frameworks that follow periodic model improvements.