The concept of artificial intelligence was first proposed in the 1950s, more than 60 years ago. However, it is only in recent years that AI has witnessed explosive growth. The main reason lies in the increasingly mature Internet of things, big data, cloud computing and other technologies.
The Internet of things enables a large amount of data to be obtained in real time. Big data provides data resources and algorithm support for deep learning, while cloud computing provides flexible computing resources for artificial intelligence. The organic combination of these technologies drives the continuous development of artificial intelligence technology, and has made substantial progress. The man-machine war between alphago and Li Shishi has pushed AI to the top of the wave and set off a new round of AI boom.
In recent years, the research and application of artificial intelligence began to blossom everywhere. With the arrival of the upsurge of intelligent manufacturing, artificial intelligence applications have been throughout the design, production, management and service of all aspects of the manufacturing industry.
Three levels of AI technology
Artificial intelligence technology and products have been tested by practice in the past few years. At present, the application is relatively mature, which promotes the accelerated integration of artificial intelligence and all walks of life. From the technical level, it is widely believed that the core competence of artificial intelligence can be divided into three levels, namely computational intelligence, perceptual intelligence and cognitive intelligence.
1. Computational intelligence
Computational intelligence means that the machine has super storage capacity and super fast computing capacity. It can carry out deep learning based on massive data and guide the current environment with historical experience. With the continuous development of computing power and the upgrading of storage means, computational intelligence has been realized. For example, alphago has successfully won the world go champion by using enhanced learning technology; e-commerce platform makes personalized product recommendation based on in-depth learning of users’ purchasing habits.
2. Perceptual intelligence
Perceptual intelligence is to make the machine have the ability of vision, hearing, touch and so on. It can structure the unstructured data and interact with users in the way of human communication. With the development of various technologies, the value of more unstructured data has been valued and mined, and the perceptual intelligence related to perception, such as voice, image, video, contact, is also developing rapidly. Driverless cars, the famous Boston power robot and so on use perceptual intelligence, it through a variety of sensors, perception of the surrounding environment and processing, so as to effectively guide its operation.
3. Cognitive intelligence
Compared with computational intelligence and perceptual intelligence, cognitive intelligence is more complex, which means that machines, like human beings, have the ability of understanding, induction, reasoning and using knowledge. At present, cognitive intelligence technology is still in the research and exploration stage, such as in the field of public security, feature extraction and pattern analysis of criminals’ micro behavior and macro behavior, development of artificial intelligence models and systems such as crime prediction, capital penetration, urban crime evolution simulation, etc.; in the financial industry, it is used to identify suspicious transactions, predict macroeconomic fluctuations, etc. There is still a long way to go to push cognitive intelligence into the fast lane of development.
02 AI manufacturing application scenarios
From the application level, an application of artificial intelligence technology may include multiple levels of core capabilities such as computational intelligence, perceptual intelligence, etc. Industrial robots, smart phones, driverless cars, unmanned aerial vehicles and other intelligent products are the carrier of artificial intelligence. Their hardware and various kinds of software have the ability of perception and judgment, and interact with users and environment in real time. They all integrate the core capabilities of a variety of artificial intelligence.
For example, a variety of intelligent robots are widely used in the manufacturing industry: sorting / picking robots, which can automatically recognize and grasp irregular objects; cooperative robots, which can understand and respond to the surrounding environment; automatic following material trolley, which can realize automatic following through face recognition; cooperative robots, which can understand and respond to the surrounding environment; With the help of Slam (simultaneous localization and mapping) technology, autonomous mobile robot can use its own sensors to recognize the feature marks in the unknown environment, and then estimate the global coordinates of the robot and the feature mark according to the relative position between the robot and the feature mark and the odometer reading. Driverless technology in positioning, environmental awareness, path planning, behavior decision-making and control, but also the comprehensive application of a variety of artificial intelligence technology and algorithm.
At present, artificial intelligence technology applied in manufacturing enterprises mainly focuses on intelligent voice interaction products, face recognition, image recognition, image search, voiceprint recognition, character recognition, machine translation, machine learning, big data computing, data visualization, etc. The following is a summary of eight AI application scenarios commonly used in manufacturing industry.
Scene 1: intelligent sorting
In the manufacturing industry, there are many operations that need to be sorted. If manual operation is used, the speed is slow and the cost is high, and the suitable working temperature environment needs to be provided. If the industrial robot is used for intelligent sorting, the cost can be greatly reduced and the speed can be improved.
Take sorting parts as an example. The parts that need to be sorted are usually not placed neatly. Although the robot has a camera to see the parts, it doesn’t know how to pick them up successfully. In this case, using machine learning technology, first let the robot perform a random picking action, and then tell it whether the action is successful picking up parts or grasping empty. After many times of training, the robot will know what order to pick up in order to have a higher success rate; when picking up, which position will have a higher success rate; when picking up, which position will have a higher success rate; If you know the order of sorting, the success rate will be higher. After several hours of learning, the robot sorting success rate can reach 90%, and the level of skilled workers.
Scene 2: equipment health management
Based on the real-time monitoring of equipment operation data, using feature analysis and machine learning technology, on the one hand, it can predict the equipment failure before the accident and reduce unplanned downtime. On the other hand, in the face of the sudden failure of the equipment, it can quickly carry out fault diagnosis, locate the cause of the failure and provide the corresponding solutions. It is commonly used in manufacturing industry, especially in chemical industry, heavy equipment, hardware processing, 3C manufacturing, wind power and other industries.
Taking numerical control machine tool as an example, machine learning algorithm model and intelligent sensor technology are used to monitor the power, current, voltage and other information of cutting tool, spindle and feed motor in the process of machining, to identify the force, wear, damage state of cutting tool and the stability state of machine tool processing, and to adjust the processing parameters (spindle speed, feed speed) and acceleration according to these states in real time In order to improve the machining accuracy, shorten the downtime of the production line and improve the safety of the equipment operation.
Figure 1 tool wear state prediction based on deep learning
Scene 3: surface defect detection based on vision
Surface defect detection based on machine vision has been widely used in manufacturing industry. Under the condition of frequent changes in the environment, machine vision can quickly identify and classify the smaller and more complex product defects in milliseconds, such as detecting whether there are pollutants, surface damage and cracks on the product surface. At present, industrial intelligent enterprises have combined deep learning with 3D microscope to improve the defect detection accuracy to nanometer level. For the detected defective products, the system can automatically make the repairable judgment, and plan the repair path and method, and then the equipment performs the repair action.
For example, PVC pipe is one of the most commonly used building materials with huge consumption. In the process of production and packaging, there are many types of defects, such as surface scratch, pits, water lines, pitted surface and so on, which consume a lot of manpower for testing. After adopting the visual automatic detection of surface defects, the surface impurities of pipes can be detected automatically by setting the minimum and maximum values of area and size, with the minimum detection accuracy of 0.15mm2 and the detection rate of more than 99%; the surface scratches of pipes can be detected automatically by setting the minimum and maximum values of scratch length and width, with the minimum detection accuracy of 0.06mm and the detection rate of more than 99%; The minimum detection accuracy is 10 mm, and the detection rate is more than 95%.
Fig. 2 surface wrinkle detection of PVC pipe (source: Weishi Zhizao)
Scene 4: product quality detection and fault judgment based on voiceprint
Using voiceprint recognition technology to realize the automatic detection of abnormal sound, find the defective products, and compare the voiceprint database for fault judgment. For example, since the end of 2018, Faurecia (Wuxi) factory has fully cooperated with the group’s big data scientists team to apply AI technology to the NVH performance evaluation (vibration and noise test) of seat angle adjuster. In 2019, Faurecia (Wuxi) factory will apply AI technology to the abnormal sound detection of angle adjusters, realizing the automation of the whole process from signal acquisition, data storage, data analysis to self-learning. The detection efficiency and accuracy are far higher than traditional manual detection. With the application of the noise detection system based on AI (Artificial Intelligence) technology in Wuxi factory, the number of personnel has decreased from 38 to 3. At the same time, the quality control ability has been significantly improved, and the annual economic benefit is as high as 4.5 million yuan.
Scenario 5: intelligent decision making
In terms of product quality, operation management, energy consumption management and tool management, manufacturing enterprises can apply artificial intelligence technology such as machine learning, combined with big data analysis, optimize scheduling mode, and improve enterprise decision-making ability.
For example, the intelligent production management system of FAW Jiefang Wuxi Diesel engine plant has the functions of anomaly and production scheduling data collection, anomaly cause diagnosis based on decision tree, equipment downtime prediction based on regression analysis, scheduling decision optimization based on machine learning, etc. By taking the historical scheduling decision process data and the actual production performance index after scheduling as the training data set, the parameters of scheduling decision evaluation algorithm are optimized by using neural network algorithm to ensure that the scheduling decision meets the actual production needs.
Scene 6: Digital twins
Digital twin is the mirror image of objective things in the virtual world. The process of creating digital twin integrates artificial intelligence, machine learning and sensor data to build a “real” model that can be updated in real time and has a strong sense of scene to support the decision-making of various activities in the life cycle of physical products. In order to complete the order reduction modeling of digital twin object, we can put the complexity and nonlinear model into the neural network, with the help of deep learning to establish a limited goal, based on this limited goal, the order reduction modeling can be carried out.
For example, in the traditional mode, the fluid and thermal simulation of the outlet of a cold and hot water pipe takes 57 hours for each operation with a 16 core server, and only a few minutes for each operation after the reduced order modeling.
Scene 7: creative design
Generative design is a process of human-computer interaction and self innovation. In product design, engineers only need to set the expected parameters and performance constraints, such as material, weight, volume, etc. under the guidance of the system, combined with artificial intelligence algorithm, they can automatically generate hundreds of feasible schemes according to the designer’s intention, and then carry out comprehensive comparison by themselves, select the optimal design scheme and push it to the designer for final analysis policy decision.
Creative design has become a new interdisciplinary subject, which combines with computer and artificial intelligence technology deeply and applies advanced algorithm and technology to design. The widely used generative algorithms include parametric system, shape grammars (SG), L-systems, cellular automata (CA), topology optimization algorithm, evolutionary system and genetic algorithm.
Scenario 8: demand forecasting, supply chain optimization
Based on artificial intelligence technology, an accurate demand forecasting model is established to realize the sales forecasting and maintenance material preparation forecasting of enterprises, and make demand-oriented decisions. At the same time, through the analysis of external data, based on demand forecast, inventory replenishment strategy, supplier evaluation and parts selection are formulated.
For example, in order to realistically control the production management cost, Honda hopes to know when the customer’s future demand will occur. Therefore, it establishes a prediction model based on the customer sales and maintenance data of 1200 dealers, and calculates the number of vehicles returned to dealers for repair in the next few years. This information is further converted into the indicators for the preparation of various parts in advance. This change has enabled Honda to achieve a prediction accuracy of 99% and reduce the customer complaint time by three times.
At present, with more and more enterprises, universities and open-source organizations entering the field of artificial intelligence, a large number of successful open-source software and platforms of artificial intelligence continue to pour in, and artificial intelligence ushers in an unprecedented outbreak period. However, compared with the financial industry, although there are many application scenarios of artificial intelligence in the manufacturing industry, it is not prominent and even develops slowly.
The main reasons are as follows
First, it is very difficult to collect, utilize and develop the data in the manufacturing process. In addition, the enterprise’s database is mainly private, the data scale is limited, and the lack of high-quality machine learning samples restricts the process of machine autonomous learning.
Second, there are differences among different manufacturing industries, which require high complexity and customization of AI solutions.
Third, there is a lack of leading enterprises in different industries that can lead the development trend of deep integration of artificial intelligence and manufacturing industry.
Editor in charge: CC