Artificial intelligence (AI) is becoming one of the most important enablers of autonomous systems. But in order to become more widely used, it has to be industrial-grade.
By Rainer Brehm
The message was startling: “Porsche’s electric Taycan overtakes the classic 911.” That headline appeared in October 2021 in the German weekly Der Spiegel. At that time, Porsche had sold 28,640 Taycan models in a year—about 700 more than the flagship 911, which the carmaker has produced in quantities reaching millions over the course of six decades and eight generations of models. The electric Taycan appeared only two years ago and is decidedly unusual for a sports car, with its electric drive, roominess, and floor covering made of recycled fishnets.
But what was most unusual was the planning and implementation of production. There wasn’t enough space at the Zuffenhausen headquarters for the new production facilities. Manufacturing had to be extremely flexible to quickly respond to changes and custom requirements. There also had to be a significant reduction in carbon emissions and resource consumption. Established methods weren’t sufficient.
So Porsche dared to take a revolutionary step: it abandoned the assembly line. Instead, mobile automated guided vehicles (AGVs) convey the Taycans to various workstations on multiple floors, based on the equipment required. The time from the initial planning of the plant to production of the first car took a mere four years.
Porsche’s innovative manufacturing is a model for future production. In all industries, smart technologies based on comprehensive digitalization are going to ensure ever-greater flexibility and shorter innovation cycles, as well as products that are more customized, manufacturing processes that are more sustainable, and an ecological footprint that’s seamlessly transparent along the entire supply chain. This change will apply to many types of products, including cars, machine tools, roller bearings, polyethylene terephthalate (PET) bottles, chemicals, and sugar cubes.
AI-Powered Manufacturing
The comprehensive automation of production steps allows machines to perform hundreds of thousands of repetitive tasks extremely efficiently, reliably, and economically. But when the manufactured products and their packaging are subject to frequent changes, current production concepts are pushed to their limits.
This is where new technologies come into play, especially autonomous self-learning systems that can immediately respond to changes and individual specifications with the help of artificial intelligence (AI). These systems depend on consistent data, sensors, connectivity that includes Industrial 5G, and the integration of shop-floor technologies in corporate information technology (IT).
The methods for acquiring and evaluating data, including on the shop floor, have had tremendous advances. Many plants, machines, and products and workpieces are generating their own data. To optimize production, this data is evaluated either in the cloud or, increasingly, on-site with edge computing.
In manufacturing, a number of AI applications can recognize and categorize specific patterns to improve productivity.
Northern Italian machine-builder E.P.F. Elettrotecnica produces systems for manufacturing brake pads. Its customers used to need trained personnel to perform quality control, because conventional image-recognition software couldn’t detect the pads’ surface structure and identify rejects, and employees needed at other stations had to assume this task.
E.P.F. developed a technology to automate quality control by connecting a camera to a dedicated AI-processing module with a neural network that could automatically assess the quality. This process initially required employees to train the digital control system and show it defective pads. The system now continuously optimizes itself.
Siemens’ Electronics Works Amberg (EWA) in southern Germany annually produces 17 million Simatic components for automating plants and machines. The automated production facilities experienced a bottleneck in their automatic X-ray inspection, where mass-produced components are functionally tested. Each fingernail-sized part had to undergo an inspection process, which slowed production.
Engineers at EWA fixed the problem using AI. Important data from ongoing production is now transferred to the cloud via the Totally Integrated Automation (TIA) environment which consists of a controller and an edge device. Experts train an algorithm that’s fed information when the quality of the soldered joints on a component is unsatisfactory. The algorithm then examines the process data collected for the component and establishes causalities. After the training phase is completed, the algorithm recognizes the probability of defects when process data deviates from the norm and it then sounds an alarm. Only then are the relevant components inspected in the X-ray machine, while the vast majority can pass through without further inspection.
Siemens has trained an algorithm to predict the probability of defects to streamline X-ray testing of printed circuit boards.
Getting AI to Industrial-Grade
These examples demonstrate that AI can significantly boost the efficacy and efficiency of industrial processes and serve as an important enabler on the path to the factory of the future. But despite these initial successes, AI needs to become industrial-grade—robust, reliable, and trustworthy enough to run mission-critical processes on it—before its use in industry can become widespread.
A lot of requirements still have to be met. Critical production processes require quality-assured AI development processes and the seamless traceability of autonomous actions performed by AI-supported components. The AI also has to be resistant to all types of faults.
AI projects also depend on intensive collaborations between AI experts, automation specialists, and industry experts. The only way for industry to use the potential of the new manufacturing environment and make AI the enabler of new business models is by getting highly qualified experts from different sectors working together, including through partnerships with customers, suppliers, service providers, companies outside the industry, scientists, startups, and even competitors.
These kinds of collaborations result in complex business ecosystems in which digital enterprises like Google, Microsoft, and Amazon, with their massive IT resources for cloud computing, can make a substantial contribution to creating models for AI and machine learning (ML), to the models’ training, and to the development of scalable solutions.
We also need intelligent marketplaces whose members can offer their expertise, goods, and services, such as production capacities, raw materials, and production knowledge. AI can bring all these elements together, coordinate supply and demand, and serve as a sort of digital general contractor for pooling and controlling the providers’ individual services, including payment processing and shipping.
Nevertheless, given the increasing shortage of skills and the growing complexity of production, it will be extremely important to not lose our focus on the human factor. AI is based solely on statistical information. Whenever there’s a need for creativity, control, application, training, or troubleshooting, employees will always take precedence. AI systems need to be as simple and intuitive as possible for users so they don’t overwhelm the people directing the technologies.
If we succeed in making AI comprehensively fit for industry, the technology can reach its disruptive potential and make grand visions a reality, such as connecting partner companies to build any product to a consumer’s exact specifications.
In the automotive industry, vehicles already integrate many semi-autonomous solutions, including lane-keeping assistance, adaptive cruise control, and parking assistance systems. The Porsche Taycan offers an intersection assistant that can warn about obstacles and engage the braking system.
Fully autonomous sports cars may not be far off. They’ll be manufactured in autonomous factories where employees no longer have to perform monotonous manual tasks but instead will serve as choreographers in making highly customized and climate-neutral cars.
The success of automation technology has always been linked to its simplicity, allowing customers to program it with minimal training and without the need for specialized IT expertise or external service providers. Future autonomous systems will also be measured against this. The greater its simplicity, the sooner businesses and consumers will enjoy all the benefits of this technology.
Learn how Siemens can help your organization integrate AI and other future technologies for a leap in productivity.
Rainer Brehm is CEO of factory automation at Siemens Digital Industries.