As supply lines begin to stabilize and manufacturing companies come to terms with the “next normal” of new logistics complexities and freight costs, there’s a renewed focus on risk mitigation and efficiency. How can businesses improve operations to increase product yield or production line throughput and, in turn, boost their bottom line? 

Emerging digital manufacturing technology such as robotics and machine learning offer the potential to significantly impact operational efficiency. The precursor to meaningful machine learning is data collection. 

The challenge lies in enabling these efficiencies and hinges on the collection and understanding of manufacturing data at scale. Where are current processes enabling greater efficiency, and where are they hindering performance? 

Manufacturing data collection often focuses on the beginning and end of operational processes—what comes out compared to what goes in. 

Consider a company that produces fruit snacks for kids. Armed with information about how much fruit it takes to produce a single package, data extrapolation is seemingly simple. By comparing the amount of fruit used to the number of packages produced, organizations can determine if manufacturing processes meet expectations. 

The problem? Products don’t exist in a binary state. Instead, they move through multiple processes along the production line that modify their shape, alter their consistency, and prepare them for packaging. As a result, there aren’t just two steps to measure but thirty-seven distinct processes, any one of which can significantly impact overall output. Inefficiencies (loss or waste) in any step can cost your company time and money, reducing total ROI. 

In isolation, bits and pieces of data may point to a process that isn’t working as intended; if packaging tools are taking longer than expected to deliver finished products, it may suggest the need for fault analysis or preventative maintenance. 

However, manufacturing data tells a story at scale, complete with a beginning, middle, and end. In our example above, the story starts with a simple data point: the amount of fruit purchased and fed into processing systems. The endpoint is similarly straightforward—how many fruit snacks were produced?  

Analyzing the story determines if the expected yield matches operational predictions. If the numbers add up, no action is necessary. But if there’s a discrepancy between material in and product out, where do companies start looking for the issue? Is it a yield problem or a throughput problem? 

This is the value of getting the whole story—understanding how each step is influenced by the one before and impacts those after. Consider our fruit snack production line. Misalignment of machinery designed to crush and compress supplied fruit could result in excessive removal of material, in turn leading to smaller output yield. Meanwhile, unexpected temperature fluctuations in shaping or packaging tools could cause product shrinkage, again limiting total yield. 

Without the middle of the story, data points from every process in the manufacturing life cycle, addressing and remediating issues becomes a matter of trial and error, costing companies time and money. 

So, how do companies enhance their manufacturing process with relevant and real-time data?   It starts with recognizing that this isn’t an overnight solution. Unpacking operational issues across multiple manufacturing processes starts with implementing a robust Enterprise Resource Planning (ERP) system/Manufacturing Execution System to collect critical data points from multiple sources. 

If we don’t know which data points are relevant yet, then over-collect and analyze. Businesses need analytics tools to evaluate collected data and draw actionable conclusions to help automate the manufacturing process. Here, the story takes shape—while process assessments in isolation provide point-in-time insights, they’re effectively single sentences in the larger story that can be easily taken out of context. 

Let’s return to our example one final time. While single-process analysis might suggest machinery misalignment as the cause of lost product, a more holistic approach that considers the entire manufacturing process might reveal multiple miscalibrations rather than a single physical fault. Now you can decide with knowledge and metrics how to solve the problem.  

By getting the whole story, businesses can understand manufacturing processes’ interactive, iterative nature to pinpoint and remediate root causes rather than addressing operational symptoms out of context. Yield or throughput—the approach to solving these problems is different. Without data and analytics, it’s just a guess. 

Machine learning (ML) and artificial intelligence (AI) are no longer an undefined fantasy. Microsoft, Oracle, and others now make desktop ML tools. ML uses data you have already collected to analyze process efficiency.  

For example, if an incorrect material or feedstock was accidentally placed in a manufacturing line, ML tools could provide immediate alerts and potentially save substantial scrap and rework costs.

The goals of manufacturing technology application include enhanced throughput, improved yield, and increased customer satisfaction. Achieving these goals requires access to accurate, reliable, and timely data that reflects current operations and provides actionable insight to drive the next steps.