#Artificial Intelligence (AI) definitions vary widely in production in the field of industrial automation and in everyday life outside of the laboratory.
"Artificial intelligence" refers to a science that encompasses several different technical and engineering disciplines, including machine vision, computer vision, machine learning, and deep learning. When a system based on this combination of technologies is designed properly (from application analysis to final validation), it can add tremendous value to the factory.
John McCarthy, a professor of computer science at Stanford University, is known as the "father of artificial intelligence." Artificial intelligence can be defined as “the science and engineering of making intelligent machines, especially intelligent computer programs.” It's about using computers to understand similar tasks in humans, but artificial intelligence doesn't have to be limited to biologically perceptible methods.
In this case, artificial intelligence can provide manufacturers in different industries with a valuable tool for automated inspection of machine vision systems. Within artificial intelligence there are subsets of machine learning and deep learning. Machine learning uses techniques whereby machines can “learn” to improve at different tasks. One such technique is deep learning, which uses artificial neural networks, such as convolutional neural networks, to simulate the learning process of the human brain.
Deep learning, a subset of machine learning, and machine learning have become popular in the field of industrial automation due to its ability to "learn" from the continuous analysis of models over time. The process of deep learning starts with data. For example, to help machine vision find product defects, manufacturers will create a preliminary data set by uploading images describing defects or features that must be detected along with "good" images. Deep learning comes by collaboratively labeling a preliminary dataset, training the model and validating the results using test images of the original dataset, testing performance in production, and retraining to cover new cases or features.
When all factors are considered and the appropriate steps are followed, the software provides value when implementing deep learning tools into new or existing automated inspection systems, including defect detection, feature classification, and assembly verification. Task. Specifically, this software provides value in many applications such as tasks such as defect detection, feature classification, and verification verification. AI technology can help with subjective inspection decisions that would otherwise require manual inspection. AI technology can help identify scenarios that have a high level of complexity or variability that makes it difficult to identify specific features.
The application of artificial intelligence in applications does not benefit every application, and it is not an independent technology. Rather, AI technology represents a powerful tool in the automated inspection toolbox that can be deployed in several different industries, giving manufacturers a variety of options when choosing a solution. They can code solutions in-house using frameworks like PyTorch or TensorFlow, purchase off-the-shelf solutions, or select application-specific AI-enabled products or systems.
There are several off-the-shelf AI solutions on the market that allow end users to build their own models without being tied to a specific application. For example, Elementary’s QA platform offers what the company calls a “full-stack vision system” with imaging heads and machine learning software with advanced analytics designed to identify issues, continuously improve and unlock new capabilities for a variety of manufacturing processes. opinion. The system combines traditional machine vision tools, such as barcode reading and optical character recognition, with machine learning capabilities to add external inspection capabilities to the system. Overall, the system provides additional detection capabilities.
Mike Bruchanski said: "Artificial intelligence is not magic, it can't do everything, but it can add powerful new capabilities to automated detection systems. Anomaly detection - such as finding an obvious lump in the grain - — is a clear example of a machine learning-based vision tool that can work in conjunction with machine vision systems for quality control." Brushansky said common inspection applications for Elementary vision systems include consumer packaged goods. (including labels, caps and accessories), medical devices, automotive parts and assemblies, and food and beverage products (often involving unique assembly inspection versions).
He said: “In an inspection of prepackaged breakfast sandwiches, for example, it would be difficult to build a pattern that would allow the software to understand whether the cheese is not in the right place or not at all, but our machine learning tools allow the vision system to Look at stacked sandwiches to make quick decisions. Our platform provides a similar approach in medical device assembly inspections, while also performing a range of automotive inspections, from regulatory label identification to inspecting welds for dents, voids or cracks. ”
A number of application-specific artificial intelligence products have emerged in recent years with the goal of streamlining and simplifying certain tasks. In some cases, this may involve having an entire system up and running within hours. Rapid Robotics' Rapid Machine Operator (RMO) is a prime example of such a system. Each RMO is designed to handle common machine operator tasks and includes a 6-axis robotic arm, 3D depth sensor, gripper and a control box for edge computing and artificial intelligence processing. According to the company, rmo is equipped with pre-trained artificial intelligence algorithms.
Juan Aparicio, Vice President of Product at RapidRobotics, said: “Each RMO is designed to meet the customer’s unique production requirements. These modular work cells allow manufacturers to expand automation quickly, cost-effectively and with low risk. ."
Aparicio said advances in artificial intelligence are making robotic automation easier and more efficient to deploy than ever before.
"In our field, one of the most important value propositions of AI is the diversity of talent for automation. The common refrain is that automation has penetrated U.S. manufacturing. Through our work, we find that this is certainly the case Not so."
He added: "To the surprise of researchers, a recent MIT report on the future of work found that robots are rarely present in small and medium-sized manufacturers."
Aparicio said there are many opportunities for AI-based robot deployment, including quality inspection, autonomous mobile robots, assembly and generative design.
In the field of robotics, Photoneo uses artificial intelligence methods in its automation solutions to identify, pick and classify mixed types of items. The company uses CNNs trained on a large object dataset to identify items of various shapes, sizes, colors or materials. If the software encounters an object it has not seen before, it can identify and classify the object based on similar objects it has encountered or been trained on before. Additionally, the software can be trained on a specific data set if a customer needs to pick out anomalies or customization items that may cause model performance to degrade.
Photoneo PR specialist Andrea Pufflerova said: “Customers often require a robotic item picking system that can identify, pick and sort items of various shapes, sizes, colors or materials.” Integrating artificial intelligence into such a solution , enabling customers to localize and process mixed object types, including organic products such as fruit or fish. She added: “This may even include items that are often difficult to identify, such as bags that are flexible, deformable, full of wrinkles and irregularities. ”
Comprehensive Custom Solutions
Prolucid CEO Darcy Bachert explains: “As a systems integrator, our focus is on applying advanced computer vision and AI-based models to aid complex manufacturing inspection applications, as well as various non-manufacturing customers including nuclear and medical. “Our typical approach is to use computer vision or other existing tools to solve the problem in the simplest way possible. If we come across an application for which these are not a good fit, then we look at AI as an option and start by looking for off-the-shelf models that fit the specific use case, such as anomaly detection or feature classification. Bachert pointed out that open source platforms such as TensorFlow, which come with pre-trained models designed for relevant use cases, as well as the entire Python ecosystem, have had a significant positive impact on the adoption of artificial intelligence in manufacturing and other applications.
He explained: “Developing a model from scratch can be very time-consuming, which is often impractical for manufacturing customers. "However, if pre-trained versions can be utilized, then it greatly simplifies the initial investment."
Looking ahead to the future of artificial intelligence in manufacturing
The future of artificial intelligence in manufacturing, automation will It may involve using advanced analytics to identify defect trends early and ultimately prevent them from occurring. For example, machine learning can identify when a company produces more defects during certain times of the day, or when date code labels start to fade because a printer is low on ink. According to Bruchanski, the technology will identify when a process is going bad and send commands to the system or operator to make adjustments.
Pufflerova believes that the development of hybrid AI models that combine model-based and AI-driven approaches also offers potential for industrial applications.
She said: "Today it may not be enough to train a system to work reasonably well on a limited set of examples - one also needs to understand its internal representation. With traditional black-box machine learning or deep learning methods In comparison, hybrid artificial intelligence models provide faster, simpler learning, and better interpretability.”
For Aparicio, it’s hard to talk about robotic automation without talking about the future of the workforce.
He said: “To the extent that artificial intelligence and automation make human roles obsolete, robotics innovation will bring changes, but ultimately it will lead to more opportunities for humans.” “For example, the deployment of robots will always require Engineers are involved as they need to coordinate various integration processes, mix hardware and software, and design a reliable system."
As software becomes the primary tool for robot training and support, these roles are likely to expand. integrated into IT. Given the speed at which these technologies are evolving, businesses may decide to partner with a vertically integrated solution provider, allowing them to focus more on growing their business while the vendor manages the robot fleet. Bachert explained that in this scenario, the robotic workforce will shift from distributed teams to a centralized approach, allowing Robotics-as-a-Service companies to take advantage of economies of scale and centralized training.
When it comes to overcoming the barriers that prevent rapid adoption of AI, Bachette concluded that AI is just another tool that can be used for industrial automation. However, he warns, "As the open source community continues to grow and more pre-trained models become available, the barrier to entry for these technologies into real-world applications will lower. This adoption will require end customers to invest in training within their teams, as Artificial intelligence presents very unique challenges that are not always present in simple computer vision or inspection applications.
The above is the detailed content of How to choose modern AI technology for manufacturing and automation applications. For more information, please follow other related articles on the PHP Chinese website!
![[Web front-end] Node.js quick start](https://img.php.cn/upload/course/000/000/067/662b5d34ba7c0227.png)
