Printer Friendly

Focusing on vision.

Machine vision is a technology with roots that can be traced back to the 1950s (see sidebar), when forward-thinking scientists and engineers first marveled at its potential:

* Robots that could "see" would be able to perform assembly operations faster, more precisely, and more consistently than assembly machines that required extensive, time-consuming setup.

* Inspection systems that could tirelessly examine thousands of components, recognize subtle flaws, and direct machines to separate good parts from bad.

* Production-based inspection systems that could examine components at production-line speeds, and direct the control of production parameters to ensure quality.

Today, those science-fiction-like visions of yesteryear have become reality as manufacturers have come to regard machine vision as an intelligent sensor that is, in many applications, an integral part of a larger control system. As machine vision provides information about the manufactured product, the control system uses the information for closed-loop control of the production process. Manufacturing yield increases rather than decreases because the closed-loop control system ensures that only a minimal number of bad parts are produced. Using this approach, industrial manufacturers benefit in three ways: quality is improved, scrap is reduced, and yield is improved.

Objective applications

One example of this modern approach to using vision is the -Eye Cut System- developed by Computerized Manufacturing Products Inc, an Allen-Bradley authorized system integrator. A vision-based process control system for the steel industry was developed to allow volumetric measurement of hot steel blooms and slabs. Previously, the steel industry would sell its product based on volume (weight) but would cut the slab based on a length measurement. Because of variations in the cross-sectional area of the slab, a consistent volume of steel was not delivered. The "EyeCut System," using multiple cameras, measures the slab's profile and computes volume by the integration of cross-sectional areas. An industrial computer uses the vision system's volume measurements to control a cutting torch, thus providing a very consistent volume for each slab cut. The system has proven so effective that the payback period for the user was approximately four months.

In another application, vision is used to gauge stampings made from metal blanks. Stamped parts are easy to gauge-typically because they can be backlit in fairly controlled environments. The resulting high contrast images can take advantage of advanced sub-pixel resolution techniques using algorithms such as edge-finding, connectivity, and normalized correlation. The resulting, very precise information is then used to detect minute changes in the press, signaling tool wear before it leads to out-of-tolerance parts.

The metal industry also uses vision systems to control heat treating. Infrared imaging is a growing market within the machine vision industry. By analyzing the infrared fight being reflected by a hot part, it is possible to gather information about the distribution of heat across the part. Few technologies other than vision offer the precision and real time control necessary for this type of process, which can be used to monitor and control the heating or cooling of a precision gear or cog. Critical attributes of steel are developed by precisely controlling the heating and/or cooling of a specific part. This implementation often requires the use of custom control concepts to anticipate the temperature of various places on the part. The vision system helps control the process via analog output to the furnace controller.

Subjective vision systems

To this point, we have discussed machine vision systems in objective applications. That is, in applications that require vision to check the dimensions or orientation of a target object (i.e. is the target within a thousandth of the specified dimension, is the object rotated properly and right-side up?). But what about subjective measurements, such as how smooth is a surface, how much waviness is in a surface, and how closely a part's color matches specifications? These are frequently referred to as subjective measurements because, until the late 1980s, these quality standards were subject to a human inspector's judgment.

For example, in the automotive industry, exterior sheet metal panels can be marred by distortions from the draw operation, waviness from imperfections in incoming sheet stock, dings caused by dirt in the die, and defects caused by mishandling. Plastic interior panels can have waviness induced by the die surface, process setup variables, or even material selection. The rapidly expanding use of composite materials for exterior skins used on high performance aircraft has created a growing concern for control of surface waviness.

Traditionally, a variety of techniques have been used for inspection of surfaces in various industries. Yet, the inspection process is often unsupported by practical, low cost inspection methodology and lacks meaningful, quantitative analysis and documentation.

In the automotive sector, examination of exterior panels for defects is often carried out by inspectors aided by fluorescent-lamp reflections off the surface of a body panel. The inspector visually determines the amount of distortion in the reflected image of lights from the surface. in other cases, a skilled operator passes his hand over the panel's surface to feel for defects. Again, a method prone to a great deal of subjectivity and requiring extreme operator sensitivity and training to produce meaningful results.

Through the use of machine vision gray-scale analysis principles, images can be captured, stored, and analyzed. Product quality can be quantified through the use of inspection algorithms provided with vision systems that quantify defect severity.

In one system, provided by Diffracto Ltd, an operator selects a zone to be quantified through the use of a mouse, and numeric severity ratings are provided. Because of the broad variance of criteria in various industries and segments, calibration of the system quantification is usually user-provided by collecting data from a number of components, obtaining management concurrence of accept/reject criteria, and establishing inspection parameters for acceptance and process control.

At audit stations, a special proprietary screen, which reflects light back to a camera, and an image capture package are the basic vision system components. The imaging package contains a solid-state camera with integral quartz halogen light source.

Defects that might go undetected when viewed under normal light become apparent when using machine vision. Defects such as waviness near the door edges, ripples in the gas tank door, and dents in the rocker panel are made obvious even to an untrained observer. These defects, though hard to detect without machine vision, could be objectionable to a discriminating customer, even though they are in many cases less than 0.001" high.

Vision based CMMs

Vision-based inspection systems have become so accurate and reliable that they are now beginning to augment or supplant more traditional quality tools such as coordinate measurement machines (CMMs), to transform academic discussions of total quality control into reality.

Optical CMMs look like traditional touch-probe type CMMs but use cameras instead of touch probes. Videometrix's PRISM system measures large, heavy parts and small components with overall repeatability of +0.000 02" for each axis (X,Y,Z) independently.

According to Videometrix, the PRISM system typically inspects parts at a rate 10 to 100 times faster than traditional methods, without operator judgment affecting the results. Users report consistently reliable readings, even on continuous two and three-shift schedules.

The equipment operates at a speed that permits statistical data to be immediately fed into the manufacturing process to insure that bad parts are rejected. Statistical process control (SPC) software included in the system's computer provides the user with tabular and graphical printouts.

Less traditional looking systems from companies like Perceptron offer users many capabilities that on-line CMMs provide. They enable users to measure product variances from nominal at line speeds, quantify variances, and collate resulting data into useful and timely information on process stability.

Dubbed quality measurement stations (QMS) by Perceptron, the laser-based vision systems consist of three basic components: DataCam noncontact sensor, image processor, and host computer. The DataCam sensor uses structured laser light or incandescent lighting to measure in one, two, or three dimensions and can be fixtured to produce a dedicated measurement system, robotically manipulated for flexibility, or handheld for accurate manual inspection.

A QMS can gauge critical dimensions with typical allotted production cycles, and as many as 255 measurements can be recorded in less than 30 seconds.

Results from the triangulation measuring techniques used by the sensors are processed, evaluated, and compared to stored acceptable workpiece limits. Appropriate signals (accept, reject) can be related to an operator, system controller, or computer to initiate a desired response.

Critical to SPC methodology, the QMS host computer and advanced software use collected data to support and produce a broad spectrum of current quality control information. These methods include simple pass-fail sorting, histograms, X-bar and R charting, predictive analysis, Taguchi loss function analysis, and closed-loop manufacturing.

IMAGE PROCESING PAST, PRESENT, AND FUTURE

Many feel that image processing grew out of research conducted by the Department of Defense in the 1950s. DOD scientists first used image processing to scan and analyze satellite pictures. In the '60s, image processing research moved to universities where artificial intelligence techniques were first applied in an attempt to help manage the millions of bits of data that were scanned 30 times-per-second.

Industrial use of vision did not occur until the 1960s when General Motors first installed vision systems for guiding robot arms used in semiconductor manufacturing. GM had identified 44,000 potential applications for machine vision in its factories alone. Still, rooms full of computer hardware were needed to perform even simple vision-based tasks. Until computers became more powerful (through the addition of more memory, faster processors, and artificial intelligence) and less costly, vision technology was primarily restricted to labs.

The lure of immediate labor savings and quality improvements drove the early demand for vision equipment. However, many of these early systems failed due to misconceptions on how to apply machine vision in a factory environment. Failures typically were associated with one or more of the following elements:

* Sensitive and fragile equipment was unreliable or would not operate in the factory environment.

* Equipment would not tolerate normal manufacturing variations such as part orientation, surface or texture nuances, lighting changes, and temperature.

* Vision algorithms and programming were too inflexible for the application and would not allow easy changeover to new parts.

* Human interfaces were too difficult for factory floor operators.

* Equipment was too expensive to justify for production.

Until the 1980s, it was fair to say that machine vision was still a sophisticated technology in search of an application. By 1984, however, when companies like Ford and GM each announced equity participation in a number of small vision companies, interest began to grow. An industrial robot made the cover of Time, and analysts predicted that machine vision sales would well exceed $1 billion by 1990.

Dollar consumption for machine vision equipment not only exceeded $1 billion by 1990, but exceeded it by a wide margin. In its report, The US Market for Commercial Image Processing Systems (#A2310), Frost & Sullivan found that the dollar consumption for machine vision equipment reached $1.45 billion in 1990 and expected it to grow to about $6.8 billion by 1994.

Vision is no longer a technology in search of an application but, instead, a technology that offers a reliable and affordable solution to many of industry's assembly and inspection challenges. End-users are coming to realize that machine vision is not just a sophisticated technology, but a cost-effective technology offering productivity gains simply too great to be ignored. This is especially true for discrete part and process automation where machine vision uses real-time data for analysis, control, and display.

Paralleling the rapid growth in machine vision equipment are those related industries that support the industry such as: consulting services, software developers, and manufacturers of input and output devices.
COPYRIGHT 1992 Nelson Publishing
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 1992 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Stovicek, Don
Publication:Tooling & Production
Date:Feb 1, 1992
Words:1938
Previous Article:Manual CMMs for accuracy and value.
Next Article:beyond JIT and TQM lies flow management.
Topics:


Related Articles
Non-contact metrology.
Leadership in the nineties.
Machine vision focuses on quality.
Exploring our future. (ASAE Up Front).
Focusing on vision: through an environmental lens.
The need to be right.
OMRON Opens New R&D Facility in Shanghai, Capping Off Major Investment in Chinese Market; Collaborating with Universities to Make New Breakthroughs...

Terms of use | Copyright © 2016 Farlex, Inc. | Feedback | For webmasters