Every day, businesses generate massive amounts of data at the edge and store it in the cloud, using this data to rethink how to transform all their processes. To better unlock the potential of data and drive faster, smarter decision-making, companies in manufacturing, energy, mining, transportation, and agriculture are leveraging new machine technologies to optimize a variety of workloads, including engineering and design, production and asset optimization, supply chain management, forecasting, quality management, smart products, and machines.
From operational efficiency to quality control, and beyond, companies are using machine learning to transform industrial production processes in four ways:
General Electric: Enabling Predictive Maintenance for Equipment
Continuous equipment maintenance is a major challenge for many industrial and manufacturing companies. Historically, most equipment maintenance has been either reactive—repairing equipment after it breaks down—or preventative—avoiding failures through regular inspections. Both are costly and inefficient, while the best solution is predictive maintenance. Companies can anticipate when equipment will require maintenance, but most lack the personnel and expertise to develop such solutions.
Fortunately, leading suppliers in power generation equipment, solutions, and services, such as GE, have already achieved predictive maintenance for equipment. Companies don't need to possess machine learning or cloud-related technologies themselves; they simply need to use end-to-end systems employing sensors and machine learning to detect abnormal fluctuations in machine vibration or temperature, thereby receiving alerts.
These technologies enable GE to rapidly update information using sensors, transforming time-based maintenance into predictive and prescriptive maintenance through real-time cloud analytics. As the system scales, GE can remotely update and maintain sensor arrays without physical contact.
ThunderSoft: Solving Product Anomaly Detection
Ensuring product quality is just as important as ensuring equipment operation. Visual inspection of the production process typically requires manual labor, which is not only tedious but also inconsistent. To improve quality control, industrial enterprises are looking to adopt computer vision technology to increase the speed and accuracy of defect identification. However, enterprises still face many complex challenges when building, deploying, and managing visual anomaly systems based on machine learning technology. Now, enterprises can use high-precision, low-cost anomaly detection solutions that process thousands of images per hour to discover defects and anomalies, identifying images that deviate from the baseline so that enterprises can take further action.
Recognizing this trend, ThunderSoft, a globally renowned provider of intelligent operating system products and technologies, has integrated Amazon SageMaker, a leading global machine learning service, into its Smart Industrial ADC (Automatic Defect Classification) system. This helps manufacturing customers easily acquire AI-powered quality inspection capabilities in industrial production. With Amazon SageMaker, customers can build, train, interpret, inspect, monitor, debug, and run machine learning models within a unified interface, eliminating the need for complex machine learning deployments. In the implementation of ADC systems in the electrical industry, Amazon SageMaker has helped end-users reduce one-time investment costs by 42%, software development workload by 39%, system deployment time by 50%, and system operating efficiency by 35 times compared to traditional inspection methods, overcoming the obstacles to deploying ADC systems in industrial scenarios.
Swedish home-cooked food manufacturer Dafgards has also applied computer vision technology to the production process of its subsidiary brand, Billy's Pan Pizza. Billy's Pan Pizza is a microwave pizza, and the production line can bake and package two pizzas per second. Dafgards had previously installed a machine vision system, which was successfully used to detect the proportion of cheese on the pizzas. However, the problem was that this function would fail when there were too many toppings on the pizzas. By adopting a new machine learning technology based on computer vision, Dafgards easily achieved cost-effective detection capabilities. Following its successful application, Dafgards plans to expand the application of computer vision to more types of pizzas, as well as other product lines such as hamburgers and omelets.
BP: Improving operational efficiency
Many industrial and manufacturing companies are looking to leverage computer vision technology to improve operational efficiency. Typically, businesses manually monitor and audit factory sites via video to verify facility access, inspect shipments, and detect leaks or other hazards. However, this process is not only difficult but also error-prone and costly. While companies can upgrade existing IP cameras to smart cameras for better processing power to run computer vision models, this remains expensive and problematic; even smart cameras may not achieve the required high accuracy and low latency. In fact, companies can apply computer vision technology to existing local cameras using hardware, or even build new cameras using software development kits, allowing them to run computer vision models at the edge for greater efficiency.
Global energy company BP is planning to deploy a computer vision system at 18,000 service stations worldwide. They plan to use computer vision technology to automate the movement of fuel trucks into and out of facilities and to confirm the completion of valid orders. The technology can also alert workers to potential collision hazards, identify foreign objects in dynamic isolation zones, and detect oil leaks.
Foxconn: Optimizing the Forecasting Supply Chain
Modern supply chains are vast networks comprised of manufacturers, suppliers, logistics providers, and retailers. They require sophisticated methods to understand and meet customer needs, while simultaneously adjusting to fluctuations in raw material supply and external factors such as holidays, events, and weather. Failure to accurately predict these variables can lead to significant cost increases, resulting in over- or under-allocation of resources, wasted investment, or poor customer experiences. To anticipate future scenarios, businesses are leveraging machine learning techniques to analyze time-series data, providing accurate forecasts to reduce operating expenses, improve efficiency, ensure greater resource and product availability, deliver products faster, and lower costs.
Foxconn is the world's largest electronics manufacturer and technology solutions provider. During the COVID-19 pandemic, Foxconn employed machine learning technology to address unprecedented challenges in customer demand, supply, and production capacity fluctuations. Foxconn developed a demand forecasting model for its factories in Mexico to generate accurate net order forecasts. Using machine learning models, they improved forecast accuracy by 8%, expecting to save $553,000 per factory annually, while minimizing labor waste and significantly improving customer satisfaction.
To fully realize the application potential of machine learning in industrial environments, industrial products, logistics, and supply chain operations, an increasing number of enterprises are looking to adopt machine learning technologies to make production processes simpler, faster, and more accurate. By combining real-time data analytics in the cloud with edge machine learning, industrial enterprises are steadily turning their aspirations into reality, while simultaneously driving the arrival of a new generation of industrial revolution.
