• Written by Azhar Aliyar, Director

A production planner at a discrete manufacturing plant is deciding this week how to allocate production lines to meet changing demand, and they currently base it on manual checks of the production schedule, maintenance records, and inventory levels. However, to make more informed decisions, they need to query the plant's data in real-time, which requires a certain level of enterprise data readiness for AI in manufacturing.

Manufacturing complexities

The nature of manufacturing, particularly in discrete and process plants, introduces unique challenges that make it harder to achieve data readiness. Changeover losses, where production lines are halted to switch between products, can lead to significant downtime and affect production schedules. Additionally, multi-plant comparability is often difficult due to differences in equipment, processes, and local conditions. The ageing asset base of many manufacturing plants also poses a challenge, as older equipment may not be equipped with modern sensors or data collection systems, making it harder to integrate with newer systems.

These complexities require careful consideration when designing and implementing data systems, as they can significantly impact the accuracy and reliability of the data. For example, a plant with frequent changeovers may need to implement more advanced scheduling systems to minimize downtime, while a plant with an ageing asset base may need to invest in retrofitting older equipment with modern sensors.

Field-level meaning and documentation

One common misconception is that column names in a database are sufficient documentation for the data. However, this is often not the case, as column names may not accurately reflect the meaning of the data at the field level. For instance, a column named "temperature" may not specify whether it refers to the temperature of the product, the equipment, or the environment. This lack of clarity can lead to misinterpretation of the data and incorrect decisions.

To avoid this, it is essential to provide clear and concise documentation for each field, including the unit of measurement, data type, and any relevant context. This documentation should be easily accessible and maintained regularly to ensure that it remains accurate and up-to-date. By doing so, users can ensure that they understand the meaning of the data and can make informed decisions.

Systems and scope

When implementing data systems, it is easy to overlook certain systems or scope them out of the project, only to realize later that they are crucial to the overall process. For example, a plant may have a separate system for managing inventory, which is not integrated with the production scheduling system. This can lead to partial answers and inaccurate decisions, as the production planner may not have access to the most up-to-date inventory levels.

To avoid this, it is essential to take a holistic approach to data system design, considering all relevant systems and their interconnections. This may involve integrating multiple systems, such as enterprise resource planning (ERP), manufacturing execution systems (MES), and supervisory control and data acquisition (SCADA) systems. By doing so, users can ensure that they have access to a complete and accurate view of the data, enabling them to make informed decisions.

For more information on how to design and implement effective data systems, users can visit our website via this link.

Gaps and inaccuracies

Stockouts, backfills, and other gaps in the data can also teach a model the wrong lesson, leading to inaccurate predictions and decisions. For instance, if a plant experiences frequent stockouts of a particular material, the model may learn to predict that the material is always in short supply, even if the stockouts are due to external factors such as supply chain disruptions. This can lead to overordering or underordering of the material, resulting in wasted resources or lost production.

To avoid this, it is essential to identify and address gaps and inaccuracies in the data, rather than simply ignoring them or working around them. This may involve implementing more advanced data validation and cleansing techniques, such as data profiling and data quality metrics. By doing so, users can ensure that their models are trained on accurate and reliable data, enabling them to make informed decisions.

Limitations and alternatives

While achieving enterprise data readiness for AI in manufacturing can bring significant benefits, it is not a solution for every problem. In some cases, the costs and complexities of implementing and maintaining a data system may outweigh the benefits, particularly for small or medium-sized enterprises. Additionally, some plants may have unique requirements or constraints that make it difficult to achieve data readiness, such as highly customized products or rapidly changing production schedules.

In such cases, alternative approaches may be more suitable, such as using manual data collection and analysis methods or partnering with a third-party provider to access specialized data services. For more information on how to determine the best approach for a particular plant or organization, users can contact us via this link.

Ultimately, the key to successful implementation of data systems in manufacturing is to carefully consider the unique challenges and requirements of the plant or organization, and to design and implement systems that meet those needs. By doing so, users can ensure that they are able to make informed decisions and drive business success.