How to Use AI to Optimize Manufacturing Costs
Brian Hoey - January 19, 2021
According to a recent McKinsey study, manufacturing and supply chain are the two functions with the greatest likelihood of experiencing cost reductions as a result of AI, with more than 60% of respondents in each area reporting savings due to AI. For anyone who’s been paying attention to the hype around Industry 4.0 in general and AI, machine learning (ML), and Big Data in particular, this probably doesn’t come as too much of a shock. These technologies thrive on large quantities of information, and the average production plant produces gigantic caches of data each year. The result is that there’s plenty of information for AI and ML algorithms to analyze in pursuit of efficiency gains and cost savings.
As the folks at Supply Chain Dive noted, these cost savings usually come in the form of improved "yield, energy, and throughput." In other words, manufacturers are able to leverage analytics technology to do more with the capacity they already have, reducing downtime and improving performance. This is, broadly speaking, due to two important factors: network optimizations and forecasting improvements. Basically, in AI-enabled environments, planners are able to more effectively see what’s coming in terms of things like demand and capacity restrictions—and, at the same time, they’re better able to work within constraints to get finished products out the door on time. The result is improved unit costs and better use of capital commitments.
At a relatively high level of abstraction, analytics-driven forecasting and optimization are the keys to reducing manufacturing costs. But how does that actually play out in a real-life production plant? For starters, it takes the form of detailed scheduling for manufacturing runs. Where a human planner simply eyeballing the situation—or even using a digital planning solution that doesn’t have analytics capabilities baked in—might have trouble creating the optimal path forward considering all of the many constraints that will apply to any given production program. This includes things like labor and machine capacity, customer requirements, parts and raw material requirements/bill of material needs for individual products, and much more—all of which creates a complex web of possibilities that can be difficult to cut through.
Luckily, AI is designed to tackle exactly this kind of complexity—which means that a planner with an AI-powered solution can take on detailed scheduling, even in a job shop environment where a single, optimal answer would take too long to achieve. This puts you in a position where you can structure a given workload on an incredibly minute, detailed level in a much shorter period of time than ever before. From there, you can become much more responsive to both incoming and projected orders, improving your on-time performance, reducing waste and idle time, and delighting your customers in the process.
Predictive Machine Maintenance
One of the most powerful uses for forecasts in an AI-driven technology ecosystem is forecasting future demand in order to improve your S&OP process. In this way, you can more effectively match demand and capacity to keep the costs associated with overages and shortages at bay. But once we move beyond that use case, we can also look at some of the more niche ways in which AI has the power to transform the production floor. For instance: predictive machine maintenance.
Right now, unexpected machine idle time (due to breakdowns or unplanned maintenance) is a costly fact or life for most manufacturing businesses. But if it were suddenly possible to see breakdowns coming in advance and take proactive measures to deal with them, you could cut out all of the costs associated with unplanned idle time (e.g. increased unit costs, premium freight for rushed orders after the plant is online again, etc.). With AI-integrated into your software ecosystem, it’s possible to do exactly that. Predictive algorithms take in live production data from your factory floor, and over time they correlate specific factors (factors that would be virtually impossible for a human planner to identify without AI assistance) with impending breakdowns. Once those links are established, planners can get notifications when a breakdown seems likely, such that they can take proactive steps to find the least disruptive moment to perform maintenance.
Here, it might be useful to call out the distinction between cost reduction and cost optimization: the former is what it sounds like—simply taking measures that reduce existing costs in some way—while the latter involves changing and updating systems or processes with an eye towards organically decreasing costs over time. Something that makes premium freight unnecessary can help you cut costs, but a system that makes your supply chain more lean can offer cost optimizations, i.e. by reducing your capital commitments for inventory. The power of real-time monitoring is that it can power both: the former by way of reduced downtime, and the latter by way of more complex, real-time optimizations.
Speaking of cost optimization by way of optimizations: one of the other most important ways in which AI can help you reduce costs—and keep them down—is through production network optimization. Like we saw with the other examples above, this involves algorithmically identifying patterns that the naked eye couldn’t possibly discern, but this time with the intention of finding less wasteful and more efficient ways to structure a production network. The more efficiently your network is structured, the more effectively you can keep logistics costs down and reduce the extra work that comes with re-planning when a disruption or other supply chain snafu arises. Plus, you’re able to get leaner as you go and thereby reduce your capital commitments going forward.
Often, this kind of optimization will occur via the use of digital twins that offer simulated models of your supply chain. These take in real-time data so that the model constantly reflects real supply chain conditions, and they give you a collaborative environment in which to test out theories and verify them using AI-powered models. As it happens, many manufacturing and supply chain businesses are also utilizing some kind of supply chain twin for planning purposes—which means that the biggest obstacle that many have to face in terms of optimizing costs is simply scaling this technology throughout their different touch points and use cases. This may take some nominal investment, but generally speaking it’s worth it.