What are some of the emerging applications of artificial intelligence and machine learning in MES?

With the advent of IoT and the Cloud, it is now possible to integrate machine data into MES, use MES data for contextualization to predict failures, and do predictive maintenance. The architectural approach for doing this is to have a system at the edge to collect the data locally from machines without exposing data sources to the internet. Build and train the machine learning model in the cloud and then deploy it at the edge or in the cloud to get the inference. The exact location of deployment depends on the use and latency requirements. A system that generates maintenance orders like MES can use these inference results to schedule maintenance and assign priority.

By leveraging AI and ML algorithms, manufacturers can analyse vast amounts of data and identify patterns or anomalies that might go unnoticed by human operators. For example, detecting subtle changes in machine behaviour that could indicate an upcoming breakdown or failure. This predictive maintenance capability not only helps prevent costly breakdowns but also extends the lifespan of equipment, resulting in significant cost savings. With technological advancements in these fields, new exciting opportunities and use-cases are opening up to further exploit process data, beyond the current capabilities of MES.

Traditionally, Quality inspections were carried out manually or through disconnected machines. The results of those inspections were then fed to MES either through point interface or manually to report the final good product count. Automated inspection can help remove this disconnect between execution or inspection and reporting manufacturing results. The AI/ML inference results can be directly fed to MES, resulting in faster corrective and preventive action. The main advantage of using AI/ML for automated Quality Inspection is that it helps increase the frequency of quality inspection, reducing errors and costs associated with the human element. Integration with MES is just an added advantage with an integrated inspection system.

Essentially, it closes the analytics gap from post production to real time production. I’d rather know that I’m out of spec as soon as possible to minimize waste/rework. It comes at a cost though. It takes a ton of data and attention to get that tuned for any given product. Best applied to high volume, quality critical item production.

A comprehensive quality model can also now be considered. Predictive quality based on vendor lots, human resources, machine availability and past performance, and environmental conditions is now a possibility.

The benefit of using AI for quality control is the ability to contain a poor batch or lot in an earlier step of manufacturing to reduce downstream cost, rework and timeline delays. Machine learning has shown promise of being able to crunch multiple variables and create correlation models between quality of intermediates early in a multi-step process and the final product, thereby allowing early intervention and remediation.

MES can also give insights such as "Consider performing this check more frequently, as there are often spec breaches, or critical defects detected", or "Consider doing this check less frequently, as it take time, but there are rarely failures".

Model Predictive Control (MPC) techniques have been in existence for a few decades now, but with the usage of reinforcement learning techniques, they are becoming be a reality for process control. In terms of Process Optimization, the primary use case has been the growth of Digital Twins to be able to simulate various scenarios for process optimization and with human-machine decisions, be able to push the new recipe to the shop floor.

There are some examples available: Replacing a filter in the brewing process. Or deciding when another rework round is not usefull anymore. Companies are looking for low hanging fruit in there production process to optimize

Digital Twin use cases are to drive specific business outcomes, such as predicting the behavior of a physical system based on real-time and historical data remote monitoring. In manufacturing environments, MES data adds context to data captured from sensors, devices, and machines. For example a machine may produce data such as running speed, process parameters, and sensors might give ambient temperature and humidity. When you combine this data with MES data such what product is running on that machine, who is operator, what is the raw material and when this combined data is contextualized and imposed on immersive 3D or 2D image then it gives the complete picture to user. Then this information can be used for decision making.

A digital twin is a virtual representation of the physical world, and MES is a virtual representation of a physical factory. So in it's simplest terms MES is a digital twin, keeping track of the current and historic state of the manufacturing facility. Digital twins are also used for simulations and "what-if?" scenarios. For example, simulating how a production line would run with 1 more operator, or 1 more parallel machine on the bottleneck operation.. Or hypothesising what would happen if the tool was changes later or earlier. Data-driven decision making is crucial, and this is a tool that could help support investment decisions, and also help drive efficiencies throughout the factories.

The rise of Copilots in supervisor guidance is a clear use of how previous knowledge can be encapsulated into a AI bot that can guide and assist in supervisory decision making.

One of the primary functions of MES per MESA guidelines is to make work instructions available to shopfloor workers. A lot of times these documents are bulky, and relevant information for specific tasks is spread out throughout. Using AI/ML to extract data from the text can help to find such information quickly and improve productivity. Another example is that a warehouse or receiving clerk wouldn’t need to screen a lengthy document to locate the part number, lot/batch number, or any other relevant information to enter into the system. These documents come with material, and not all vendors supply the info in the same form. However, due to various compliance, users must capture this information before using the material.

The ability to apply generative AI to address the operator training pain points is a clear winner here. It is no secret that shop floor operators are usually afraid of fast-pace change in their work environment. They frequently lack enthusiasm for training in new systems. Where AI can help take the stress away is through an AI Assistant that has 100% context awareness and can provide guidance on the next steps at any point in the process execution. I have implemented such AI assistance for my customers. It assists newer, less-experienced operators during the shift changeover and on night shifts with knowledge of the production recipe and everything that happens in the manufacturing process. These operators are taken care of really well.

Per the ISA-95 model, MES sits on the border of IT-OT, and traditionally, due to latency sensitivity, they are hosted on-premises instead of distant corporate data centers. Now MES can also be hosted in the cloud to take advantages of cloud such as cost, scalability, faster innovation and purpose built services . Businesses can mitigate security concerns using cloud-native services for data encryption, identity and authentication, regular auditing of data sources feeding into MES, and other best practices around data governance and handling.

What about Planning and Scheduling with AI. I have heard of several companies doing innovation projects (proof of concept) around this. AI is probably also usable in cases where the optimization is very complex. But the requirement is still do it automated.

By comparing historical MES data (e.g. real cycle times) with current target times in the master data, patterns can be identified which are the basis for AI algorithms. Therefore, production scheduling is for sure one of the MES areas which can benefit the most of AI. Better target times lead to a better planning which enables a smoother production and a higher on time delivery rate.

MES has revolutionised the way manufacturers operate by providing access to vast amounts of accurate, real-time data direct from the shopfloor. This data is a goldmine for companies looking to leverage Artificial Intelligence (AI) and Machine Learning (ML) algorithms for advanced analytics and predictive modelling. Advanced algorithms like these are only as good as the data and hypothesis they are built on. MES provides the data - so is absolutely crucial to these initiatives.

Production Planning and Detailed scheduling is definitely a key area that can be powered by AI algos using MES real-time execution data from all stations (from stamping-painting to final assembly) in mix with key inputs around 3Ms (man, machine, material) to optimize production efficiency and handle complex variants mix. PPDS allows adaptability to changes in demand with dynamic rescheduling and lead time reduction which is a huge factor for auto OEMs scalability, also brings in crucial $$ savings through lower inventory and labor optmization.

While this is not an exact MES use case, the growth in generative design can help speed up manufacturing decision making. When alternates have to be considered at a plant level based on materials available for that shift, a quick redesign of the product and process within approved specification reduces the risk of scrap, rework and poor quality.