Monthly Archiv: March, 2014

Inspection Systems

When we think of lab automation systems, we think of systems that are capable of inspecting elements but their suitability to carry out such tasks depends on a number of factors. Someone once asked me whether or not a robot would be capable of carrying out the final inspection of experimental product. This proved to be a difficult question to answer, as it is unclear whether automated machines can be decisive when evaluating the final stage of an experiment. In fact I would say that at present this technology is certainly not capable of such feats but who knows what their capabilities may be in the future.

Inspection System

I would still trust a human operative over a robotic one when it comes to making these kinds of decisions. Perhaps in situations where there a few obvious anomalies with the final product, a detection system or inspection device could be used to quality assure the product. In manufacturing environments, such a system could be used to detect parts that were missing from the product or broken pieces that had been damaged during the manufacturing process. These tasks are highly suitable for this system and there are many benefits to using inspection systems in this way. The process is normally carried out by human operatives who seem to fall short when executing this task, often missing defects or anomalies. This is why this role needs to be carried out by robotic systems and this is also why companies have spent so much money developing this technology.

Inspection Systems

The question needs to be answered; why is it human operatives are unable to adequately carry out inspection work? It seems to be the repetition of this task causes the human brain to effectively tune out, making it increasingly more likely they will miss any defects or missing parts. In a similar way to how a word processor works, a machine can perform a search of a document to find a single word in a few seconds. This is a task that may take a human operative hours but there are no guarantees that the word will be found. A human operative has difficulty fully focusing when the task is of a repetitive nature but a machine can handle this task without any issues. Even if a human being could manage to achieve this task, the role would be so boring that they most certainly would quit. This is before taking into account how hazardous such a job may be to the operative, with eye strain being a common issue.

Robotic Perception

Following the identification of a window of interest, analysis is carried out to target certain features in the lab. Threshold analysis is often described as transcribing an image to black and white pixels. This not a new approach by any means but it is still considered one of the most affective forms of analyzing images. The main reason why this was so popular in the infant days of machine visioning was due to its ability to screen out the sensing systems operational variability.

Robotics

The threshold approach is used in many binary systems, though these systems are not limited to a singular threshold method. If you consider the problem of identifying a snake which is grey in colour, on a black and white tiled floor for example. This approach would simply portray the image as being a black snake on a white floor, with no tiles visible at all. Maybe the automated system in question does not need to worry about this but other systems will and therefor a solution was needed. In a laboratory setting, two thresholds are set select the gray part of an image and allow the device to pick-up just that part. For this very reason, threshold systems are designed to accept two distinct thresholds, allowing the system to get around the problems outlined above. If only one threshold is required, the other threshold can be tuned to one extreme of the brightness scale, leaving that threshold completely void.

The act of finding the perfect threshold value for a binary visioning system is a key process in creating a successful vision automation system. This problem is much trickier than you may think, with many external variables affecting the outcome. Selecting a threshold that lays exactly half-way between the two light extremes can lead to a totally dark image, if both thresholds are below mid-range. Some sort of intelligent approach must be adopted to select the required threshold. Where this intelligence comes from is entirely up to the designer of the project but some sort of manual input from a human operator may be required. Although computer programs have improved greatly in how they calculate these ranges, they are still not up to the task completely. This does, however leave a system that is not fully automated and therefore cannot be classed as an automated device as such.

As you can see, there are many issues with this approach but most of these can be overcome with a little planning and design.