• Written by Azhar Aliyar, Director

A production planner checks the schedule for the week, noticing that a changeover is required on one of the lines to accommodate a new product run, and wonders how this will affect the overall production targets, which are tracked in the enterprise resource planning system, and also queried through AsscherAi.

Manufacturing complexities

The manufacturing environment presents unique challenges for AI systems integration, particularly in discrete and process plants where production runs to a tight schedule. Changeover losses, where production is halted to switch between products, can be difficult to account for, and may require custom integration with the manufacturing execution system to accurately track downtime. Additionally, multi-plant comparability can be an issue, as different plants may have different systems and processes in place, making it harder to compare performance across sites.

The ageing asset base of many manufacturing plants also poses a challenge, as older systems may not have the same level of connectivity or data output as newer systems, requiring creative workarounds to integrate with AI systems. This can be particularly problematic when trying to integrate with systems that are no longer supported by the vendor, or where the original documentation is no longer available.

Read and write paths

When integrating AI systems with manufacturing systems, it is essential to consider both the read paths and write paths. Read paths are relatively straightforward, as they simply involve extracting data from the systems of record, such as the enterprise resource planning system or the manufacturing execution system. However, write paths require more care, as they involve updating the systems of record with new data, which can have unintended consequences if not done correctly.

A common mistake is to assume that write paths can be treated the same as read paths, but this can lead to data inconsistencies and other problems. Instead, write paths require careful consideration of the underlying system architecture and data models, as well as the potential impact on downstream systems. For example, updating the bill of materials in the enterprise resource planning system may have a ripple effect on the manufacturing execution system and other related systems.

Reconciling definitions

One of the biggest challenges in integrating AI systems with manufacturing systems is reconciling definitions that differ between systems. For example, the definition of a "batch" may vary between the enterprise resource planning system and the manufacturing execution system, which can lead to confusion and errors when trying to integrate data from these systems. To overcome this challenge, it is essential to establish a common data model that can be used across all systems, which can be a time-consuming and complex process.

Most people get wrong the idea that reconciling definitions is a simple matter of mapping one system's definitions to another, but this approach can lead to problems down the line. Instead, it is necessary to take a more nuanced approach, considering the underlying business processes and rules that govern each system, and working to establish a common understanding of key concepts and definitions. This may involve working with subject matter experts from each system to ensure that the definitions are accurate and consistent.

Latency budgets and precomputation

When integrating AI systems with manufacturing systems, it is also essential to consider latency budgets and what can be precomputed. Latency budgets refer to the maximum allowable delay between the time data is generated and the time it is available for use in the AI system. In manufacturing, latency budgets can be very tight, particularly in process plants where production is continuous and delays can have significant consequences.

To overcome latency issues, it may be possible to precompute certain data elements, such as aggregations or summaries, which can then be stored in a cache or other intermediate storage system. This can help to reduce the latency associated with querying the systems of record, and can also help to improve the overall performance of the AI system. For more information on how AsscherAi can help with latency budgets and precomputation, please visit our website.

Limitations and wrong choices

While AI systems integration can be a powerful tool for manufacturing organizations, it is not a silver bullet, and there are certain situations where it may not be the best choice. For example, in situations where the systems of record are highly customized or proprietary, it may be difficult or impossible to integrate with AI systems. Additionally, in situations where the data is highly sensitive or regulated, such as in the pharmaceutical or aerospace industries, it may be necessary to implement additional security measures or controls, which can add complexity and cost to the integration process.

In these situations, it may be better to consider alternative approaches, such as manual data entry or custom scripting, which can be more straightforward and less prone to errors. If you are unsure about whether AI systems integration is right for your organization, please contact us to discuss your options and determine the best approach for your specific use case.