Machine Learning in Manufacturing: Moving to Network- Wide Approach
By Paul Boris, CIO - Advanced Manufacturing, GE
Up until this point, machine learning in the Industrial Internet has focused on optimizing at the machine level. We have access to a ton of data about machine function and productivity that we have used to run our machines at full capacity for as long as possible and predict many maintenance issues.
But now it’s time to take the next step and start looking at network-wide efficiency. By moving beyond the nodes of machine data and analyzing the bigger pic¬ture, manufacturers can unlock the true poten¬tial of machine learn¬ing. Network-focused machine learning al¬gorithms will include data sets like inventory, material cost and labor cost, machine capability and performance – fac¬tors that have been considered on a plant-by-plant basis already. However, by opening up the entire network’s worth of data to these network-based algorithms we can unlock an endless amount of previously unattainable opportunities.
With the move to network-based machine learning algorithms, engineers will have the ability to determine the optimal workflow based on the next stage of the manufacturing process. We already have the ability to run machines at extremely high productivity rates, but what’s the point of stressing a machine if the next piece has been delayed for two weeks? Machine learning algorithms will give plant engineers the knowledge that they can run certain machine at a slower to reduce the wear on the equipment, while still completing its output in time for the next stage in the manufacturing process. The engineer needs the authority and the ability to move in and amongst the data, letting the algorithms understand the impact of the current performance on the next action and recommend a course to the operator that most effectively meets the business objectives.
The Gig Economy
Looking beyond the machines themselves, machine-learning algorithms can reduce labor costs and improve the work-life balance of plant employees. By utilizing more data from across the network of plants and incorporating seemingly disparate systems, we can better enable the “gig” economy in the manufacturing industry.
Looking beyond the machines themselves, machine-learning algorithms can reduce labor costs and improve the work-life balance of plant employees
For example, you might employ a very specific skillset based on the products you build or machines you run. Using advanced data and machine-learning algorithms you may have identified that the likelihood of mechanical issues or production disruption is imminent. Instead of having the specialized labor arrive either too early to be fully productive or too late to avert the issue, an organization can be more prescriptive as to when and where they deploy key resources if at all. And while many companies do this now with seasonal or surge labor, we’ve seen that this model can be utilized effectively in many of the new consumer-based business models that are emerging. A shorter work day that provides the same amount of productivity for both the worker and the plant is a win-win, it’s the theory of working smarter not harder.
Today, large manufacturers often have plants set-up based on industry or product set. For example, they have one plant focused on healthcare products and one focused on aviation. By enabling machine learning to look across the entire network, manufacturers will be able to more effectively move to a multi-modal facility production model. What part, machine or skill profiles are similar across gas turbines and jet engines, for example, and how could one site be tuned to fulfill demand for multiple businesses? Enabling the seamless flow of data (the Digital Thread) is critical in this case, but machine-learning algorithms can determine that the most cost effective production strategy is to make 1,000 parts in Kansas and ship 300 to the healthcare plant in Wyoming and 700 to the aviation plant in California. By moving to a multi-modal production model and analyzing a broader, real-time data set, the capacity of each plant is optimized to increase the efficiency of the entire network.
Across a diverse manufacturing operation, at any point in time there will always be some plants that have excess capacity, while others are struggling to deal with spikes in demand. Today, many manufacturing plants are siloed, and are forced to determine how to maximize their own operational effectiveness even if that includes planned downtime or overtime. By sharing data across the network, a plant could better share any excess capacity or shed workload to better optimize the supply network, as opposed to a single operation. While this is done in a macro sense today, the window of opportunities will continue to shrink as we approach a real-time supply chain for the most complex, engineered-to-order products.
Schedule for Purpose
When producing extremely large or complex products, scheduling production to optimize cost and delivery can be difficult on both the manufacturer and the customer side. Every manufacturing supply-chain executive can share horror stories about customers requesting to move up or push back their delivery dates and the chaos that can ensue. In the future, the algorithms will be able to provide the ability to schedule for purpose. When one customer says that they want to move their order back from March to May because their facility won’t be ready, the algorithm will determine whether the production schedule can be adjusted to incorporate another customer’s request for an expedite, or if the delay might be of other benefit to the facility like shedding overtime planned to meet the original demand, maximizing both employee and machine productivity.
These are all opportunities that can, and at some level, are being realized right now – just not in the most complex manufacturing operations. The technology exists, but industry must first move from applying machine learning to haphazard array of node-based data on machine or cell performance to looking at the network as a whole. The real value in machine learning is in the algorithms that tell us where we should be investing in people, tools, techniques and technology across the entire manufacturing network as informed by real operational data, rather than what to do with any single machine.