Quality Control AI-based technologies help in quality control
AI-based technologies such as Machine Learning and Deep Learning are transforming the way manufacturers are looking to solve their quality control problems for their respective parts or moulds. In fact, the Deep Learning technology is gaining attention from across the industry due to its numerous advantages. Read on to know more…
We often read that next-gen technologies such as Artificial Intelligence (AI), Machine Learning (ML) and Deep Learning (DL) are gradually being incorporated into today’s smart factories that are on their way to become ‘Industry 4.0’ compliant. But what’s the difference between these terms? Let’s find out.
What are AI, ML and DL all about?
Artificial Intelligence, simply put, is all about smart machines or systems that are capable of carrying out tasks which are usually done by human intelligence. For example: chatbots on websites.
Machine Learning is another trending term and one that is often confused with AI, although there is a slight differentiation between them. ML is a branch of AI that helps software applications automatically learn by making use of historical data and algorithms. This process enables the system to improve from experience based on the data fed into it and then predict outcomes with high accuracy.
Whereas, the ‘Deep Learning’ technology is a branch of machine learning and consists of a neural network of three or more layers that allows it to ‘learn’ from huge sets of data thus helping it to make accurate predictions. The additional layers assist in optimizing and refining the data for accuracy. DL is used in the development of next-gen autonomous cars as well as many other routine applications.
In short, Machine Learning and Deep Learning are both AI-based technologies which boast of numerous advantages and are used across diverse industries including the manufacturing sector.
The ‘Quality Control’ Factor
Quality is considered to be a crucial factor as it can lead to huge losses, affect performance of a specific part and at the same time also be responsible for a decline in the firm’s reputation. Hence, a lot of research and analysis goes behind choosing the right quality control technology or process for one’s solutions.
There are various solutions that have added AI-based technologies to ensure that manufacturers deliver nothing less than perfect parts to their clients. An example of this is Zeiss’ Surfmax series. Equipped with machine learning which is based on algorithms that are developed in-house and optical technology, the solution can not only carry out accurate surface defect inspection on multiple parts from the automotive, aerospace, medical, and consumer electronics industry but also classifies the defects automatically and in real-time.
Elaborating on the Deep Learning technology with a special focus on quality control, Christian Eckstein, Product Manager Deep Learning Tool, MV Tec mentions, “Deep Learning offers a vast array of methods to solve quality control problems of parts/moulds. Those include the classification of images of parts into different defect classes. Objects can be detected in images and individual features or defects can be segmented. Deep-Learning-based systems are even able to find anomalies in images that differ from the ones that the system has been trained on.”
The company’s MV Tec Halcon software for machine vision is combined with the Deep Learning technology as from the firm’s experience, deep learning alone almost never solves the problem. Instead, it is usually combined with traditional machine vision to tweak the system for the best results in the given hardware and processing time constraints. Having all the tools available in one package really enables the developer to do their job efficiently.
Ruben Ferraz, Vision Sales Product Lead, Cognex says, “Processes are always susceptible to failure therefore it is important to have a system that can visually inspect all parts produced. Deep Learning technology excels when it comes to potential defects which always have different appearances, sizes, and shapes. DL is not a rules-based system, this means that it can detect defects based on examples, as a human operator does, the power of a well-trained neural network will reduce the number of false positives or negatives and therefore increase the process quality.”
The Cognex Deep Learning software is developed uniquely for factory automation. The focus on Deep Learning based on images is combined with a large variety of different algorithms (Deep Learning, Edge Learning, and Classical Vision Tools). The software can be deployed using sophisticated PCs or extremely easy-to-use software that is embedded in low-end or high-end smart cameras. In summary, high performance, easy-to-use (no need to be a Deep Learning specialist or a programmer), and a lot of flexibility in terms of deployment platforms are the key advantages of the software.
In addition to this, the start-up firm Covision lab has introduced a revolutionary software that automates and scales visual inspection as well as defect detection. Equipped with Machine Learning and computer vision, the neural networks-based solution helps one to achieve an accuracy of up to 99 %. About 30 parameters can be incorporated within one processing cycle and new ones can be added later. Once the changes to the detection and classification processes are finalised, manufacturers can accelerate production volumes. With this solution, costs associated with quality control can also be significantly reduced.
AI-based Technologies in Industrial 3D Printing
These technologies can be applied in the additive manufacturing industry too. For instance, the AI software company Peltarion joined hands with the 3D printing firm Amexci to develop a proof of concept around quality control in the industrial 3D printing sector, which makes use of the Laser Powder Bed Fusion (L-PBF) method. In the Laser Powder Bed Fusion method, there is a high possibility that the part or product might get deformed due to the high temperatures used in the fusing process. Hence, each layer needs to be checked for quality and therefore, each product generates innumerable images with vast amounts of data which are impossible to analyse.
In this background, Peltarion developed a deep learning model which aims to assist the firm in solving its quality control woes. The new model was expected to group similar images and then be trained to identify and predict defects but for this it required huge data to learn from. This is when Amexci launched the ‘Rosetta Protocol’ initiative in which companies within the same field were asked to share their respective image data in order to further train the game changing model for the industry.
Better than Standard Quality Control Solutions
With the rise of digital transformation amongst the industry, manufacturers are also looking at new and more technologically advanced solutions to meet their quality control requirements. This means that standard quality control solutions are now passé. “Standard quality inspection technologies have been used in the last 40 years and they are reaching their limits. Deep Learning technology opens a completely new horizon by complementing classical machine vision with neural networks that can perform inspections that were not even thought of before,” Ferraz shares, “Many inspections have a certain level of unpredictability and natural variations which belong to the process. The only way to overcome these challenges is using a technology that mimics the human brain. This is what DL does and many companies have already understood the benefits and are investing heavily in the Deep Learning technology.”
Adding to this, Eckstein elaborates, “Deep Learning has the potential to take advantage of the information hidden within a vast amount of image data. The system is given some structured data and is tasked to replicate that structure. In contrast to traditional machine vision, the features that are selected to do this are not explicitly taught but learned by the system through millions of interactions of trial and error. This can for certain problems lead to much better results. Also, problems that could not be solved in the past can now be solved with this new approach.”
He further mentions that manufacturers are keen to use the best technology to solve their problem. However, they usually are under no illusions, so it does not necessarily have to be a deep-learning-based solution. Deep Learning is still quite hardware demanding and requires a lot of data, that sometimes is not available. Also, some manufacturers still hesitate since deep learning systems act like black boxes and the decisions made by the system can seem opaque to machine acceptance engineers.
The way ahead…
On a concluding note, Ferraz opines, “Deep Learning will contribute to getting companies ready for the 4th industrial revolution (Industry 4.0). You cannot have connectivity and automation if you do not automate inspections. Deep Learning technology with its human-like intelligence enables companies to overcome the existing limitations of inspections which are done only by humans. We have the right technology to transform the industry and we have already begun”.
Although this technology is already being used in the sector, it can still be further advanced in the background of today’s ever changing technologies and market requirements. As Eckstein says, “Deep Learning has not yet reached its full potential. Constant progress in CPU, GPU and TPU (Tensor Processing Unit) technologies will make much more complex network architectures viable for industrial use. Also, we see a move towards Deep Learning in 3D. At MV Tec, we are evaluating how promising 2D technologies like Anomaly Detection can be combined with 3D vision to find surface defects that cannot be detected on 2D images only – while only having been trained on ‘good’ images.”
With so many insightful views, we can state that Deep Learning seems to be the next big technology for manufacturers to solve their quality control problems for parts or moulds. So, which DL technology are you going to choose for your business?