How smarter robots are pulling manufacturing shop floors to the future

At the ABB Smart Power factory in Nelamangala, Bengaluru, the quality inspection process is no longer manual. An artificial intelligence (AI)-enabled camera takes pictures of the finished moulded case circuit breaker (MCCB) from five sides and checks against the algorithm for defects such as terminal thickness, height, printing errors and scratches. Only when all the parameters are met—no smudged text, printing error or structural damage—is the MCCB cleared for packaging. Saju SR, Senior Vice President, Smart Power Division, ABB India, claims zero returns since the system was implemented. The system has brought a level of precision that is impossible for a human being.  

ABB is not alone. AI—from being focused largely on software functions—is finding its way to physical environments such as factory shop floors and warehouses as well. At the Tata Steel factory in Jamshedpur, an AI-enabled sensor tests for corrosion in pipes, a change from the earlier system where the operator had to decide whether to continue running the system based on the camera feed. Now, the AI model determines the extent of the corrosion and takes a call if it’s safe to run the machine. The 118-year-old company runs 650 AI models at manufacturing locations; over 500 are in India. It has also set up a remote maintenance centre to monitor mining operations. “Our Integrated Remote Operations Centres, Raw Material Supervision Centres and Integrated Maintenance Excellence Centre provide intelligent, location-agnostic oversight of processes from hundreds of kilometres away. This enables centralised predictive and prescriptive maintenance, saving over 1,500 hours in asset maintenance time,” says Jayanta Banerjee, Chief Information Officer, Tata Steel.

To be sure, smart automation has been transforming Indian shop floors for over a decade, so much so that robot-powered assembly lines and warehouses are a common sight across the country. However, physical AI goes beyond smart automation. The difference is the intelligence layer. “In smart automation, there is no decision-making by the machine. It’s a pre-programmed algorithm operating on the basis of inputs provided in the past,” says Saju. “With physical AI, you add an AI layer to physical systems so that they can have the cognitive abilities of a human being.” This gives the system the ability to understand what is happening and take a decision based on the input received. The ability to process device data with AI locally on the physical device is an essential component of physical AI, also sometimes referred to as Edge AI.

L&T Technology Solutions, which works with companies to set up digital twins, or virtual replicas of their factories, has developed an AI perception system that allows vehicles to make split-second navigation and safety decisions without relying on cloud connectivity. It has created plant-level twins being used by global OEMs to identify micro-efficiencies in assembly lines, says Amit Chadha, Managing Director & Chief Executive Officer, L&T Technology Services. “This has led to a demonstrated 10-12% productivity gain. This evolution, from descriptive to prescriptive and predictive twins, has been powered by AI,” he says.

Earlier this year, a report by the World Economic Forum and the Boston Consulting Group, Physical AI: Powering the New Age of Industrial Automation, said “recent innovations in software and hardware have ushered in a step-change in robotic capability, enabling robots to perform complex tasks in dynamic environments with simpler deployment. Advances in AI and complex simulations, enabled by accelerated computing using graphics processing units (GPUs), have made it feasible to run AI models and algorithms in real time, unlocking new applications. This AI-based approach focuses on enabling robots to perceive, plan and act in complex, real-world scenarios, effectively achieving a level of physical intelligence.”

Automation initiatives have traditionally focused on jobs that are repetitive or unsafe for humans. But the intelligence layer is reducing the human-machine interface by giving the machine cognitive and perceptive capabilities to take decisions that earlier required human intervention. “Some jobs are hazardous or require extensive training. These companies are becoming leaders in AI adoption,” says Vijaya Ghosh, Partner, BCG. When it comes to areas such as predictive maintenance, the system is trained to think like a human operator and move away from a rule-based system, which is what is happening at companies like Tata Steel. Here, the system can read cues like abnormally high temperature or unusual sounds, analyse, and decide to shut down the system for safety reasons.

The marriage between AI and automation has created a digital super operator on the shop floor. Most factories have an old-time employee with a wealth of non-transferrable knowledge about the system who knows exactly how to handle each machine. Over time, training automation systems to act this intelligently will reduce the risk of dependence on one or two people capable of effectively managing the shopfloor. “We are asking clients to use AI to pick up that knowledge and become a smart operator who’s able to do all this and take cues not only from the experience of that person but also use data to come up with insights which didn’t exist before,” says Ghosh. This is one reason why AI-driven interfaces can be worth the investment.  

A Slow Start

As with any new technology, some companies are resistant to adopting physical AI while some want to go all out and upgrade all the systems, sometimes without considering how effective it will be for them. If it’s a set process, like tightening screws, a standard automation software is likely to do the job just as well, say experts. “If it is a critical process and adds value to the customer or improves the quality of the output, AI can make a difference,” says ABB’s Saju.

The technology is still at a nascent stage and quite expensive. At the same time, labour costs are comparatively low in India, so the decision will depend on what you want to achieve. “If you want the output from the process line to not be dependent on who is working the line at the time, then there’s a case to be made for intelligent automation,” he says.

Warehouse Upgrades

Beyond the shop floor, physical AI solutions are also finding their way into warehouses and supply chains. “Technology plays a pivotal role throughout Amazon Fresh’s supply chain. At the sourcing stage, we have built an ecosystem with a team of qualified agronomists offering expertise to farmer partners,” says Srikant Sree Ram, Director, Amazon Fresh India. This improves farm yield and product quality through machine learning and computer vision-based algorithms that simplify supply chain processes and identify defects in fruits and vegetables, reducing wastage. “We also utilise shelf-monitoring cameras that feed into machine learning models to spot defects,” adds Sree Ram.

The shelf-monitoring solution includes an app called Johari, which uses computer vision models and Wi-Fi and IoT enabled cameras to identify pre-determined defects such as cuts and cracks in fresh produce. In the manual mode, an operator can submit a picture of the produce which is then analysed to highlight any item that does not meet quality standards. In the automated monitoring mode, cameras on shelves take pictures at predetermined intervals and assess for defects. The process takes six seconds.

Globally, Amazon has been among the early adopters of physical AI in its warehouses, operating the world’s largest robotics fleet, according to the WEF report. Leveraging autonomous systems that bring together mobile robots, AI-based sorting and generative AI-guided manipulators to improve fulfilment has resulted in 25% faster delivery and boosted efficiency in some locations by 25%.

Build or buy

While several companies are going for customisation, some are adopting ready solutions such as autonomous mobile robots (AMRs). Mahindra & Mahindra is using physical AI with AMRs in production units and warehouses. For example, it has set up vision cameras which facilitate internal routing of sub-assemblies in production shops. “All AI initiatives we are working on in manufacturing are aimed at the shop floor. We have developed models to derive actions and are validating the accuracy of these actions. Once validated, these AI models will be integrated into our smart automation systems to scale up physical AI,” says Nalinikanth Gollagunta, CEO, Automotive Division, M&M. The company is working on a project where AI models will predict the integrity of a spot weld in real time. AMRs are a common sight on shop floors at ABB too. These are autonomous and embedded with a 3D camera, which detects obstructions, triggering a reroute if needed.

At the M&M manufacturing facilities, the smart automation-based inter-shop conveyors work by executing a predefined routing pattern with the PLC ensuring that the set route is always followed. Using physical AI enables flexible routing based on signals received from an AI model designed to find the shortest route and depending on current material movement across the lines.

However, implementing the AI layer is not a straightforward process even if you have a robust foundation. At the ABB plant in Bengaluru, an AI-powered camera regulates turnstile access to a hardhat area where employees are required to be in personal protective equipment and safety shoes, something which wasn’t always followed. The AI-enabled camera takes a picture and does a quick assessment of whether the person is wearing the appropriate gear; only then is he or she allowed in. The system now works efficiently and has even won an internal award, but Saju says it has taken them a lot of time to get to this stage. “Training the algorithm came with its own set of challenges. For instance, the system could not initially differentiate between a hardhat and a regular cap. Or it would reject hats of a different colour. The system took time to learn but that’s part of the process,” he says.

Tata Steel’s Bannerjee sounds caution. “Fundamentally, AI is not a standalone journey. You need to first invest in Cloud, Edge and data technology. You need to build the road before the car. That has been our transformation journey at Tata Steel. We’ve invested in the Cloud and are working on integrating the data layer at all our sites in India and worldwide.”

LTTS’ Chadha also emphasises the importance of having robust data management practices to benefit from AI. He cites the example of a European industrial manufacturer. “The biggest hurdle was not AI algorithms but unifying disparate data sources from decades-old systems. Once harmonised, AI-driven twins unlocked predictive maintenance insights that reduced unplanned downtime by 15%,” he says.

For Indian enterprises, this is turning out to the among the biggest hurdles to implementing new technologies. Apart from the larger conglomerates, a majority of manufacturing facilities still use old machinery which doesn’t have IoT capabilities. Investing in newer machines and collecting and collating this data is the first step towards an AI integration road map.

Trust is another factor. AI adoption is a multi-pronged change management exercise where building trust among engineers and operators in the AI’s recommendations is as important as building a robust and reliable AI algorithm.

BCG’s Ghosh says once there is a general acceptance that AI can deliver results, the second hurdle is achieving economies of scale. “It becomes a value versus cost play. If there is a proven value, as against what you must pay, there will be adoption,” she says. Given that the technology is still at a nascent stage, the cost of the AI integration may be more than the productivity gains in some cases.

The next stage is likely to see the use of generative AI-based agents on the shop floor. Global engineering major Siemens has been investing in integrating AI tools into its entire range of offerings. “Industrial AI is the key to transforming industries. Generative AI will be a valuable tool for every engineer and an indispensable part of the future engineering process across all industries,” says a Siemens India spokesperson. This will tackle critical industrial challenges like high machine downtime, time intense failure diagnosis and response.

Companies must realise that deciding whether to bring physical AI tools on their shop floor is no longer a question of if but when. There is a shift from batch and mass production towards customisation. Manufacturing will have to evolve to meet these changing customer demands, and AI powered automation will be the most effective way to get there.