Whether you call it Industry 4.0, Smart Industry or the Fourth Industrial Revolution, the transformation taking place in the manufacturing industry in recent years takes the same form: increased networking of machinery, systems and technology to improve productivity, efficiency and quality. The industry is becoming particularly hungry for smart objects, big data and... artificial intelligence!
Artificial intelligence against non-quality
In an increasingly competitive production environment, quality is a key differentiating factor. And non-quality represents a major risk to performance and competitiveness, leading to very high costs for industries. Are you in the 20% of companies that admits they are unable to measure non-quality?
The criteria of non-quality
Even if you are able to measure non-quality, it is highly likely that your measurements only partially reflect reality. Non-quality is usually analyzed through four criteria: non-compliance itself, as well as its material consequences (rejections, replacements, etc.), financial consequences (warranty, late penalties, loss of margins and markets, etc.) and intangible consequences (customer dissatisfaction, loss of trust). For industrial companies, reducing non-quality is therefore a key issue for reducing costs and increasing performance.
AI, a new tool for quality control
In a report published in August 2017, Infosys revealed that machine learning is seen by 75% of manufacturing companies as a key factor in transformation, with the cognitive tasks managed by artificial intelligence considered a key factor by 57%. Machine learning and deep learning therefore reinforce the existing infrastructure and human resources, giving you the ability to proactively identify mistakes and faults that weaken your production chain and the quality of your products.
This is where artificial intelligence has some very strong arguments in its favour. Machine learning and deep learning algorithms today contribute to the growing automation of quality control in production chains, helping considerably reduce the number of faulty parts, and the high costs resulting from them.
Computer vision et machine learning : quality control reinvented
Complete automation continues to represent a major challenge for businesses, as quality control is still largely based on the visual ability of human beings and their - limited - ability to take into account and adapt to potentially variable conditions. Artificial intelligence, on the other hand - on the condition that it is fed by correctly "trained", good data - is not, by its very nature, restricted by physiological, occupational health and variability constraints.
Accelerating the fault detection process
More systematic and more reliable than the human eye, quality control is carried out throughout the production process, including during the selection of raw material. The CRIQ (Center for Industrial Research of Quebec) has, for example, perfected a system based on the combination of digital vision technology, sensors and mathematical models, enabling the quality control of wood shavings in the paper industry, based on their freshness, dimensions and the presence of faults and contaminants, before they reach refinery workers.
This is also taking place on the assembly line itself. In the automotive industry, the bodywork painting step is where most faults occur. Optical inspection solutions, such as the one developed by Isra Vision, today make it possible to automatically detect faults and surface irregularities, whatever color or type of paint is used. Whilst other systems, at other stages in the production chain, will be used to control the positioning and assembly of electronic components, or the assembly of seat belt buckles.
Controlling food presentation: Domino's Pizza Checker
Quality control can also take place at the very end of the line, to check a product's compliance with presentation standards. This is the case, for example, at Domino's Pizza, which has installed a "Pizza Checker" in its chains, a video control system, driven by artificial intelligence that checks that pizzas delivered to customers look the way they are supposed to.
Size, shape, distribution of ingredients... everything is checked automatically to ensure the satisfaction of even the most demanding customers. If let's say the red, green and yellow peppers aren't equally distributed across the pizza, the Peppina will end up on the rejects pile.
Control measures now go beyond simple observation, going even as far as prevention. This is called "predictive quality". A combination of digital visual inspection (computer vision) and machine learning algorithms make it possible, unlike with traditional image processing techniques, to automate complex inspection tasks and detect or even anticipate faults.
A constant need: precise and reliable data sets
Rather than relying on fallible manual inspections, you can now using artificial intelligence, machine learning and deep learning to increase the precision, efficiency and speed of your quality control process. However, because these predictive quality models need to begin by objectifying subjective criteria, it is essential that they rely on algorithms trained by human intelligence. Which is to say, fed by training data sets that are precise and relevant enough to enable the system to be truly efficient.