When fish processors demanded a higher level of automation, an Icelandic firm assembled an in-house team to engineer an X-ray inspection system.
Vidar Erlingsson, electrical engineer, Marel hf, Gardabaer, Iceland
As automation sophistication increases, more partnerships are being forged between technology suppliers. At Gardabaer, Iceland-based Marel hf, a go-it-alone preference prevails. Founded by two engineers 23 years ago as a developer and manufacturer of scales for fish processors, Marel cultivated a made-here culture as it expanded into portioning, deboning and robotic equipment for meat, poultry and seafood companies. The firm even assembles its machine PLCs from custom-built printed circuits. Its product development department employs 60 engineers.
The impulse toward in-house development was demonstrated when a Norwegian fish cooperative solicited Marel's assistance four years ago to automate the removal of pinbones from salmon. Chronic labor shortages, even with high operator wages, raised the possibility of flying Norway's catch to China and back for processing. Instead, Marel's technicians modified a deboning machine to remove pinbones and coupled it with X-ray inspection and a feedback loop to operators if bones are detected. A second-generation X-ray unit recently was introduced, and technicians continue to refine the programming to reliably identify bone in fish and poultry.
Heading Marel's X-ray project is a team of four electrical engineers, including Vidar Erlingsson, a member of the robotics and product inspection group. Erlingsson graduated from the University of Iceland before pursuing a MS in electrical engineering at the Technical University of Denmark (DTU) in Lundtofte, Denmark. He joined Marel part-time in 2000 while working on a master's thesis on calibration of a vision system, then became a full-time staffer upon graduation from DTU.FE:
Several vendors manufacture X-ray units that are being successfully applied to high-speed food inspection. Why not partner with one of those vendors?
Quality assurance X-ray is becoming more affordable, but the software usually is designed to analyze a wide assortment of objects because the units must work for a broad number of applications. Also, most quality assurance applications are in packaging; our machines are designed for a wet environment. Humidity is the biggest concern. Cameras and bulbs react to any variation in humidity. Stabilizing the equipment was a critical design consideration.FE:
Was your earlier work with vision systems useful in X-ray development?
Imaging was helpful, though nothing directly linked to X-ray. The methodologies are very different, but studying engineering involves learning to think in the right way, so in that sense it was helpful. With a vision system, the naked eye can see the object you're trying to detect. With X-ray, you can't see it, and there is a lot of trial and error involved in developing the logarithms. You really have to dig into the physics of X-ray.
We didn't know much about X-ray before we started. Perhaps 30 percent of the effort was really intense experimentation and mastering the principles. The rest involved practical issues that had to be resolved. That was the most frustrating aspect.FE:
Given the state of advanced imaging, has X-ray become a mainstream technology?
The medical field has done much to advance the basic technology, but food plants are completely different environments. They require simplified systems that can take pictures and process images at high speed, and the price tag the technology must carry is much lower. We couldn't add a lot of fat to the system. For example, shape analysis of the different areas of a fish would be nice, but that would make the machine more complex, expensive and difficult to maintain.FE:
What were the objectives at the project's outset?
Cod has about 14 pinbones, each about 1.5 inches long and 0.2-1 mm thick. The bone is attached to the skin, not the spine. The meat surrounding the bone is the loin, the most valuable part of the fillet. Workers were throwing away a lot of the loin when removing the bones, and the cooperative was bringing in foreign workers to address labor shortages. Therefore the objectives were to reduce manpower and increase yields.
We had acquired a Danish company named Carnitech. Carnitech's mechanical engineers modified an existing machine to remove salmon pinbones. The machine removes 60%-70% of the bones. The idea was to couple the mechanical system to an X-ray machine where any remaining bones can be identified. The image then is presented to operators who remove the remaining bones.
Pollock fillets with pin bones detected by the X-ray unit feed back to the operator area. About 30% of the fillets still contain a few fine bones after passing through a mechanical bone-removal machine.
What level of resolution is delivered?
The X-ray source scans the target about 1,000 times a second and builds up an image with 0.25 mm resolution, or about 1,000 pixels. A 300 mm (1 ft.) fillet being conveyed at 0.3 meters a second (60 ft. a minute) would be scanned 1,200 times.
Depending on the species being X-rayed, the parameters of the logarithm may have to be altered a bit, which isn't a problem. What can be difficult is the contrast between bone and flesh, and that is affected by the fish's diet, the temperature of the water where it lived and other factors. When X-ray is applied to package inspection, only very dense objects need to be identified. With in-line inspection of fish or poultry, there can be very little grayscale contrast between the flesh and the bone, and that can be a problem. A more sophisticated algorithm is needed.FE:
What difficulties have you encountered in the field?
The system for Norwegian cod performed well, but we faced several challenges in applying it to saithe (pollock). The orientation of the pinbones is different, which necessitated a change in the mechanical system. Our engineers designed a conveyor that becomes cone shaped as the fillet approaches a cylinder with grooves that pull out the bones. The belt resembles an inverted V when the fillet passes under the roller, which has two axes of motion. The fillet straddles the belt's centerline, and the cylinder rolls over the surface, picking out the bones.
The system was deployed at a Canadian processing plant, but performance was unacceptable. The problem was the variability of the saithe being brought to the plant. However, the system now is in use in Dalvik, Iceland, where it is detecting bone in tails, and at a plant operated by HB Grandi in Reykjavik.FE:
Is worker safety an issue?
Worker safety is a key consideration, of course, and strict rules are applied to render the machine harmless. The key statistic is the annual dosage of radiation exposure for workers. Extremely low leakage occurs, below background levels found in an environment such as Reykjavik.
Most of the components for X-ray are designed for industries where high levels of energy are used. When looking for small bones in flesh, low energy levels are used, and sourcing standard components that utilize low energy is a challenge. We continue to look for hardware that can improve the contrast levels of mass using low power.FE:
How has X-ray and automated bone removal performed at Grandi?
In the first five months, the improvement in yield was in the 1.5-2.5% range. There also has been a reduction in labor costs. Plant management estimates it will reduce manpower from 23 to 14 people on the line. Based on the beta test, a complete line was to be installed in September.
The test also was designed to demonstrate the effectiveness of X-ray inspection. In the past, the technology often failed to meet the standards that were set. For example, a UK quick-service restaurant set a standard of two bones per pound of processed fish for fish sticks and similar products. The X-ray units used in the test couldn't meet it, so there was no reason for the chain to encourage its processors to adopt the technology. If X-ray can help lower the number of bone complaints from 100 per million servings to, say, 15, the customer will champion it. Incremental improvement and a robust system that works is the goal.FE:
Is the payback from this level of technology sufficient to justify the capital investment?
Based on the Grandi trial, we estimate the payback at one year. High Scandinavian labor costs improve returns: line operators in Iceland earn about 300,000 kronnur a month ($4,300 US). Finding people to work even at those wages is a challenge, which makes automation critical.
But this technology has to catch on as a standard if it is going to be widely used. It is customer driven: if a superior product is produced and fewer consumer complaints are received, clients like McDonald's will strongly urge suppliers to install inspection systems.