Sure, the human eye is amazing, but there could be problems if it must inspect hundreds of manufactured parts per minute. So, just short of hiring the Six Million Dollar Man or Bionic Woman, what can device-makers do to ensure that all parts are of the best quality? It all starts with machine vision, say Vantage Consulting Group’s Vito Pirrera and John Jordon in this guest column.
The human eye is a magnificent thing, but it has its limits.
There has been extensive research into how best to manage human inspection, be it in the manufacturing of semiconductors, automotive parts or aircraft engines. For every manufacturing process, there is a calculated error rate that defines how many people and how often inspections are required.
In general, manufacturers know the systems they use today work – from the most simplistic viewpoint, it is obvious that the reliability and safety of almost every manufactured product has increased dramatically over the past two decades.
On the other hand, companies are in a race to improve productivity, reduce waste and increase shareholder value. It seems like every day employees are asked to do more with less; the implied threat is that if they don’t manage to do that, their competitor will figure something out and the firm will lose market share.
But one area that we see having a dramatic potential for productivity improvement is in using machine vision for high-throughput medical device parts. Like most industries, device production lines are generating more products faster and with increasingly tighter tolerances.
As a result, there’s more for humans to inspect and the margin for error is thinner.
Some examples we have seen of this simply defy reason. In one case, workers were asked to inspect thousands of parts per hour, pushing more than a hundred parts per minute. For some scenarios, this might be possible, but this specific scenario was small injection-molded parts where the defects were minute. It was impossible to be accurate at the requested inspection rate.
The reality was that this specific human inspection was actually costing more than it was saving. No real improvements were being made and the valid testing was happening downstream, when the parts were assembled.
On a fast-moving line producing hundreds of parts per hour, inspectors typically sample parts – perhaps fewer than 1% of them. The typical process for this is well known: at line start-up and at certain intervals, samples are taken and inspected using “reference” parts or pictures. Sometimes measurements are taken and documented. This process continues until the lot is complete, which can take anywhere from one shift to a week’s worth of production.
As with anything, the issue comes when an exception is found. Let’s consider an example of hourly sampling on a continuous throughput high-speed line. The system has been running for two shifts and is in the middle of the third shift. As soon as a defect is found, additional samples are taken from the line and the retained materials from previous sampling are also measured. If the defect is severe enough, production is stopped.
Now it becomes challenging: sometimes the retained samples show some defects that were simply missed. It’s possible that the entire lot is quarantined and reevaluated, all while production is halted.
From a cost perspective, this is catastrophic: production costs, quality costs, potential shipment delays, all falling under the category “cost of quality.” Obviously, that’s better than shipping bad parts – but does it have to be this way at all?
Inspecting Parts In Real Time
What if, instead, the manufacturer could inspect every part as it’s produced? In the medical device arena, we’ve found this to be both a powerful and economical solution for closely inspecting parts in real time.
Machine vision typically encompasses lighting, lenses, imaging technology, software, hardware and a human-readable display. With simple controls, the machine vision system may also be integrated very cost-effectively with existing line controls.
Alternatively, it’s possible that the system can be integrated with robots – vision guided robotics (VGR) – so analytical data can continuously optimize product output. As machine vision technology matures, it is performing ever-more-intricate inspections on faster-moving products – in many cases, on every single part or product moving down the line.
North American sales of machine vision systems and components increased 10% year-over-year to $597m in the first quarter of 2017, according to the Automated Imaging Association, the industry’s trade group. The global market for machine vision system components should reach $30.8b in 2021, according to BCC Research, reflecting a five-year compound annual growth rate (CAGR) of 10.2%.
“The intensifying need for superior inspection and increasing automation are the key influencing factors paving the way for the notable adoption of machine vision technology,” says Grand View Research. “The technology is gaining a considerable traction across food and packaging, automotive, pharmaceutical, and other industrial verticals owing to abilities, such as improved detection of objects, enhanced analysis, monitoring tolerance and accurate component measuring.”
Some of the more well-known and common vision inspections that are currently being made include reading variable data, such as human-readable text and linear or 2D codes printed on parts, devices, and surgical instruments.
And some of the more interesting medical device inspection applications that are becoming increasingly cost-effective include:
Measuring parts as they exit the molding process;
Detecting scratches or particles in injection-molded parts or on the surface of medical patches;
Verifying component orientation and part alignment during assembly;
Verifying that products in kit packaging are in the correct location and orientation; and
Reading barcodes and verifying text on flexible products such as solution bags.
Different machine vision inspection systems collect different data types from parts or products. For a product with sleek, organic curves, the system may likely collect scanned surface data. Line scan cameras can detect flaws in printed material moving at 60 mph. Data can be collected from X-ray, thermal or infrared images.
Online systems make sense for high-throughput, dedicated lines. Offline systems make sense for slower or more flexible ones.
The question is whether the benefits of machine vision are worth the costs. Here’s a process for addressing inspection challenges and determining whether machine vision is the right choice:
Analyze your manufacturing cost profile. Look at a representative period – a month, quarter or half-year – and determine where your manufacturing costs are actually going. How much product is generating revenue? How much is going to scrap? How much is being rejected by customers?
Examine returns. Which products are customers sending back? How often is this happening? Why is this happening, and how is this jeopardizing relationships with your clients? Can you quantify the cost of your human inspection errors?
Document your defects in a Pareto chart. A Pareto chart is a useful quality control tool that classifies causes of problems, quantifies their role in the overall quality problem, and compares them against other causes. The bottom line? It spotlights where you’ll get the most bang for your buck if you start eliminating causes of errors. Focus on the greatest offenders.
Tally the costs of flaws. Go beyond revenue and scrap, but quantify losses and liabilities in market share, brand equity, crisis management, legal exposure and client relationships.
Consider the online versus offline choice. Online machine vision systems inspect every part or product as it moves down the line. Manufacturers need to invest in high-quality lighting and mechanisms to ensure that products are held firmly for the imaging system, and to ensure that all the surfaces or structures of the parts are imaged. Offline systems, on the other hand, are simpler. They require an operator (or robot) to pick a product from the line and move to a table for a detailed inspection. On a fast-moving line, however, you wouldn’t be able to inspect 100% of the output with an offline system. Online systems make sense for high-throughput, dedicated lines. Offline systems make sense for slower or more flexible ones.
These are a few of the basic considerations in determining your need for machine vision as a way to improve quality and revenue. Although it will require a capital investment, that investment will be defrayed by additional efficiency, revenue, brand equity, relationship quality, and human health and safety.
And with one vision system versus many human inspectors, you’ll also be able to move inspectors to roles where the limits of human eyesight don’t jeopardize business performance – or lives.