Friday, May 27, 2016
Publication Date: 06/1/2007
ARCHIVE >  June 2007 Issue >  Special Feature:
Test & Measurement

Self-Learning Vision Systems Automatically Adjust Parameters
PC-Eyebot self-learning vision system in semiconductor fabrication. .

Many manufacturers are getting a rude awakening. Their older automated machine vision systems lack the sophisticated processes needed to test today's complex electronic equipment. And while in some parts of the world it is less expensive to use human test inspectors than to invest in vision systems, the tradeoffs are lower accuracy and inconsistent results.

It is vital that the test and inspection process be as good as it can possibly be because when products malfunction or fail in the field, the reputation of the entire company suffers. New prospects are discouraged from contacting the manufacturer and current customers defect to competitors. Sales plummet and the blame eventually falls on manufacturing.

Fortunately, there is a new generation of self-learning machine vision systems, such as the PC-Eyebot manufactured by Sightech Vision Systems, that provide a level of test and measurement accuracy previously unachievable. Based on Neuro-RAM technology, these advanced systems "learn" up to 13 million features per second and "train" themselves to adjust the go/no go parameters based on data recorded during the inspection process. This is particularly useful when inspecting and sorting electronic surface mount semiconductor components. And while self-learning vision systems are as intuitive as humans, they have none of the disadvantages associated with manual inspection.

In countries where labor is cheap, manufacturers may opt to use technicians equipped with microscopes rather than machine vision systems. The fact that this manual inspection relies on the test technician's subjective decisions is just one drawback. Any myriad of emotional or physical events can impede the production of consistent test and measurement results.

Because humans are "human," there is no guarantee of consistency. If the technician is tired, will the results be skewed? How about hungry? Distracted? Bored? In applications requiring the inspection of complex parts, such as thousands of individual solders or minuscule connector pins, tedium can easily set in, causing the test technician to overlook flaws and faults that could have a critical effect on quality.

Another disadvantage of manual inspection is that it often limits testing to a random sample of products. Invariably, some flawed products are going to get through. While using humans may at first seem to be the least expensive alternative, manufacturers must consider the negative impact of faulty products.

Old Systems Can't Keep Up
While they're better than no automated test at all, many of the older machine vision systems have shortcomings that compromise the manufacturing process. They require extensive tool deployment and programming, making setup a long and costly process. In addition, the exact parameters for test criteria such "pass/fail" and "absence/presence" must be constantly adjusted or reprogrammed — either "tightened" or "loosened" — to balance the need for high yields with an acceptable number of rejected products.

A major disadvantage of these machine vision systems is their inflexibility. Most use techniques such as measuring, template-matching, pixel counting, optical character recognition, and bar-code reading. Once the test parameters are selected, they remain the same until manually readjusted. When a large number of products fail and yields plummet, production managers lower the thresholds for failure. But when quality control managers analyze the failure reports, they insist that test systems be reset to tighter standards. Making matters worse, there are often subtle defects missed by the production manager and QC manager. Just like lowering and raising a thermostat, this vicious cycle feeds on itself and everybody loses.

The solution is a vision system that can produce highly accurate data, analyze this data, and automatically adjust the parameters that determine a "go/no-go" decision. The built-in intelligence of these sophisticated, self-learning vision systems provides manufacturers with an effective test and measurement alternative that pays for itself with improved quality and yield.

Unlike traditional machine vision systems, self-learning vision employs artificial intelligence that allows the system to inspect products much the same way that a human would — intuitively. But unlike humans, this is done with complete objectivity and accuracy, since the trainable vision system automatically develops the necessary familiarity needed to perform successful, consistent inspections.

And as advanced vision systems provide continuous "learning" and "forgetting," they hold the most relevant inspection data in a prime position. The system's automatic decision sensitivity management, inherent in a self-learning vision system, internally monitors the standard deviation. This enables the system to automatically adjust its decision threshold, thus improving quality control and preventing the vision inspection system from indiscriminately rejecting an excessive number of products under test.

Chip-capacitors are inspected for malformation and metalization defects from four different viewpoints: top, bottom, front and back, at speeds of up to 60 chips per second. .

In fact, the primary and natural result of a trainable vision system is a "go/no-go" decision. This is accomplished by the massive internal work performed automatically by the system's advanced Neuro-RAM algorithms. Previously unsolvable vision problems are often easily conquered, making the trainable vision's performance almost seem like "magic".

Quality Up, Costs Down
Besides increased system accuracy and improved quality control, a trainable machine vision solution reduces test costs. For example, no triggers or strobes are used, so there's no need for system integration of special lighting. Setup and deployment are extremely fast and easy. And process or product changes are rarely required; self-learning takes care of everything.

For example, a manufacturer can easily perform brief cumulative training so that the vision system automatically learns and accumulates any new knowledge about a process shift or product change. This can be implemented as a single "learn" button placed on the manufacturing line and set up to learn for a preset period each time an operator pushes the button.

Chip-capacitors are fed through vision system with a vibratory feeder. Faulty ones are blown off the glass disk with a small air-jet. .

In addition, some trainable vision systems such as the PC Eyebot offer modes that take machine vision into new areas of image processing. One example is coloration — an innovation that takes color detection from a spectral to a "shape of color" understanding. Applications for color detection abound. For example, in this mode, an advanced vision system can detect the presence of lead or other undesirable material during fabrication. It can be trained to recognize unique texture parameters — perfect for inspecting fiber or fabric.

Best of all, a self-learning vision system can be fully demonstrated to the company's decision makers in a single visit. The trainability feature enables a variety of trials to be performed in record time, unlike the older vision systems that often required full installation to conduct an effective trial.

Cross-Industry Applications
Manufacturing is unquestionably changing the 21st Century continues to unfold. New pressures — such as increased competition in a global market, a demand for more complex products, shortened delivery times, and lower production cost requirements — will continue to force manufacturers to examine every aspect of production.

PC-Eyebot inspects hard disk platters to determine if placement is correct; in this case, an extra notch is detected. .

Self-learning vision systems empower manufactures to quickly and easily improve the quality and speed of test and measurement during the inspection process. As a result, they can address many of the concerns and issues affecting cost, quality and production rates. Some examples of applications suited to self-learning vision technology include:

  • Verify that a crimp (or notch) has been put on a metal nut as required.
  • Inspect for misplaced hard disk platters in the manufacturing cassettes for the hard disk industry.
  • Inspect the lead bonding in semiconductor components.
  • Detect bad welds in electronic components, such as the tip of a light bulb.
  • Ensure connector pins in military products are not bent, misplaced or missing.
  • Verify that automotive valve stems have slots properly placed for the accompanying retainer ring.
  • Check to ensure all bolts are in place before final assembly of any product.
  • Ensure placement of metal rods before blowing in liquid material to create foam auto seats.

Today's manufacturing environment demands smart, fast and efficient inspection. Self-learning vision technology is available now to enable manufacturers to enjoy a highly streamlined, effective manufacturing floor that helps them achieve demanding goals and objectives.

For more information, contact: Sightech Vision Systems, Inc., 6580 Via del Oro, San Jose, CA 95119 408-282-3770 fax: 408-413-2600 E-mail: Web:

search login