Many top gear manufacturers are adopting machine vision systems as a means of seamlessly yielding higher productivity, as the system can operate inline with existing manufacturing processes.

First-tier gear manufacturers are now embracing machine vision technology to streamline production lines, which, of course, ultimately leads to significant cost savings. However paradoxical it may seem, this is possible while actually improving product quality. Top gear makers are attracted to the ease of operation, agility, and flexibility in inspecting various parts that can be randomly located on even a moving surface. Gear makers can literally view first-hand how vision systems are affecting regulation compliance and accountability through a 100-percent traceability feature.

Figure 1: Gear under 360-degree inspection; one of multiple cameras shown

What Is Machine Vision?

Machine vision is a system that is able to acquire one or more images using an optical non-contact sensing device that is capable of processing, analyzing, and measuring various characteristics, so decisions can be made quickly. It is completely non-destructive, and a subset of machine vision systems are capable of real-time processing as well as inline applications.

Machine vision systems are used to meet the common needs for processes requiring vision to locate the position of objects carried through by means of a conveyor system.

Commonly called online inspection systems, objects are examined and transferred according to the inspection requests fed into automated systems for rejection or are carried through to the next stage of the process. In a defined visual process, the location of the gear in front of a camera is first identified, and then other inspection tasks are performed to make sure this object is qualified and accepted as a valid part, based on some definable manufacturing criteria, without unacceptable defects or manufacturing faults.

These robust vision systems replace human inspection with online inspection and low false reject rates. The vision system is supported by a common automated software platform, electronically connected with analog or digital cameras. This technology eliminates defective parts from the manufacturing process according to various out-of-design parameters or set conditions.

Figure 2: Chipped tooth, camera 1

How Does This Differ from Human Inspection?

Human vision is best for qualitative interpretation — highly subjective decision-making skills using opinions and reasoning-based knowledge — of complex unstructured scenes. It is inconsistent and unreliable, and 100-percent inspection efficiency is unachievable. Effective area coverage by a human is from 50 to 70 percent.

Machine vision is best for quantitative measurement of structured scenes. It serves as an intelligent information system for qualitative interpretation with rule-based thresholds and parameters.

Given that this is the environment in which gears are normally manufactured, vision systems are ideal for improving upstream and downstream manufacturing operations as each gear receives 100-percent inspection and only faultless products are delivered to the customer.

How Do Vision Systems Streamline Manufacturing?

With the robotic parts handling mechanism, a high-performance defect analysis solution is now available to gear makers who need a flexible system to increase their lean manufacturing practices.

How Can Vision Systems Yield Higher Productivity?

In-line agile and intuitive systems can perform multiple operations simultaneously. With the intelligence of visual information supplemented to the robots, the automation process adapts to varying conditions to transfer parts received to any convenient spot in the production line precisely and to perform defect discrimination simultaneously.

Where the use of single vision systems to inspect multiple gears was once an aspiration, today they are fully operational in gear plants around the world. Grippers can now automatically change over with the simple click of a recipe selection button on the human machine interface (HMI).

Figure 3: Hairline crack, camera 2

At first, users might be skeptical about machine vision products because they feel that the science of visual interpretation is hard to learn and difficult to use. However, there is an “intelligent” approach that will allow users to set up the visual inspection requirements by dragging and dropping training tools via an interactive interface without using a scripting language.

Fine-tuning can be performed using parameters collected during production. Saved images are retrieved and run to see if they are good or bad simply by transferring to a laptop.

How Do Vision Systems Offer Full Traceability?

Machine vision systems today provide solutions that enable manufacturers to meet regulatory requirements. One such requirement is product tracking. These applications are driven by global regulatory and enforcement policies that demand due diligence toward 100-percent inspection, thus ensuring product security and traceability from initial production to delivery.

Machine vision systems provide data that facilitates surface defect awareness enabling better accountability for every part that is manufactured. Traceability reports are easily generated.

Work-cell design and standardization now offer industries a wide range of functionalities in defect analysis and gauging measurements in manufacturing lines within the same production cycle, leading to major improvements in productivity, yield, and consequently, better quality and lower costs.

Figure 4: Missing tooth, camera 3

Why Haven’t Most Gear Manufacturers Adopted Vision Systems?

Concern #1: Ability to identify ambiguous objects to extract key features for defect analysis

Vision systems have now reached a point where inspection tools and image processing algorithms are able to identify ambiguous objects and extract key features for defect analysis. After many years of algorithm design experience, robotic guidance, and ambiguous parts validation, tools can cope with almost all environmental changes and variations.

Degraded patterns by means of several algorithms — i.e., morphological image processing, domain-based descriptors, and fuzzy set theory for logical parsing — are functionally represented and recognized. Algorithms are trimmed to meet the expected levels required for accuracy and processing speed and those for general object shapes and robustness against distortions due to environments with inconsistent lighting. Also, uncontrollable transfer mechanisms are applied to discriminate background noise and intelligently determine the best likelihood of classes when ambiguity dominates the clusters population. Only the most effective algorithms can be included in developing robust software tools.

Concern #2: False rejects

Manufacturing lines are shut down when they produce false rejects, slowing down production. Many false rejects are due to environmental variations including lighting, dust, oil, dirt, or grease. The solution to this is to compare all parts against red rabbits that serve as an example of rejects for benchmarks. These are recorded in the system, and fine-tuning is performed against these images.

Concern #3: Difficult to operate

Machine vision systems are indeed a new technology to many in the gear industry, but are easy to use. Since they are a new technology, a good vision system company will offer a three- to six-month warranty and service that includes online support or field visits.Online service support has eliminated the expense of both travel and prolonged downtime to reach customers both locally and abroad and to provide instant service to help monitor, diagnose, and tune-up parameters of service systems within minutes.

Figure 5: Tool-changing

Case Study

Figure 1, Figure 2, Figure 3, Figure 4, and Figure 5 are specific examples of how Abraham Innovation Systems, a machine vision system specialist based in Ontario, Canada, worked with its gear customers in using simultaneous inline inspection by a single vision system.

Visit Abraham Innovation Systems at IMTS 2016, Booth #E-4063, for a live demonstration and to discuss questions regarding vision systems.

Previous articleQ&A with David Senkfor
Next articleRedin Production Machine
received his Ph.D. in computer science from the University of Sussex, England. He has taught courses in machine vision and pattern recognition, and he led the university R&D team, implementing product prototypes at the Hong Kong Polytechnic University. He was also an adjunct assistant professor at the University of Waterloo and offered his consulting service to ASM Pacific Technology. With years of experience in machine vision and image processing algorithms, software development in manufacturing, and object identification and tracking in military applications, in 2002, Poon became the founder of CanaVision Technologies and Abraham Innovation Systems Inc.— where a wide spectrum of visual inspection and robotic automation solutions are sold to Fortune 500 companies worldwide, including those in the pharmaceutical, automotive, and aerospace sectors. For more information, please visit