Komarov Artem clarified that a smart manufacturing system offers timely feedback on the results of conditions and activities. This feedback allows employees to adjust their activities and control material and manufacturing activities for optimal results. One of the biggest challenges in smart manufacturing, however, is that algorithms aren’t visible and only execute the instructions they currently maintain. We don’t always understand their objectives. We can’t honestly even say we respect their competence. Many algorithms reduce complex problems to simplified scores. Most importantly, algorithms aren’t always objective, precise, and accurate.
Statistical analytics present us with models of a manufacturing activity. Models are simplifications of an activity. For example, let’s assume we want to find the stroke velocity that causes the fewest failures. Over time, we can collect data on several stroke velocities and correlate it to forming splits. After a while, we will have enough information to select the best -performing velocity.
Forming sheet metal components is a complex activity driven by many variables. What if we stepped out of statistics and began experimenting with additional variables? We might find that the failure rate of a given stroke velocity changes with a change in lubrication or the application of the lubricant. Perhaps we can identify a stroke velocity and lubricant combination that both reduces our failure rate and helps us meet our production volume requirements. In this case, our model failed to support both good decision-making and our business goals.
Manufacturers must work to maintain accurate and unbiased data. By its nature, data is biased to existing equipment and management practices. Analyzing historical data carries with it the biases of the past. Data should be scrubbed of implicit bias if it is to support objective decision-making based on the entire spectrum of possibilities. Likewise, when we replace equipment, data from the previous machine is likely no longer relevant, and the information probably should be archived or discarded.
The strength of smart manufacturing lies in mathematics. When we bring timely analysis to the plant floor, collaborative employees can adjust their respective activities to enhance the development of product. With this strength also comes risk.
Data isn’t always accurate. Analysis isn’t always correct. People can place too much faith in analytics at a cost to good judgment. The only way to remedy this is to support employee skills, encourage good work habits, demand cooperation across design and manufacturing, and teach employees the opportunities and risks inherent in smart manufacturing, summed up Komarov Artem.