Unusual sounds from machines such as rattling noises from air conditioners and squealing car brakes, warn us that something is wrong. Can the same apply to devices used by industrial companies such as those in the manufacturing, maritime, oil and gas sectors?
Leon Lim, CEO and founder of Groundup.AI, believes so. This is why the Singapore-based company developed proprietary Internet of Things (IoT) sound sensors and an artificial intelligence (AI) platform that can detect abnormal sounds from machines to prevent unplanned downtime.
Machine failure leading to unscheduled downtime is a key concern for industrial companies. According to last year’s report by machine health management company Senseye, Fortune Global 500 manufacturing and industrial firms are taking a near US$1 trillion ($1.36 trillion) a year financial hit due to unplanned downtime. Those companies are losing 3.3 million hours of production time annually to machine failures, and the US$864 billion economic impact of these lost hours is equivalent to 8% of annual revenues.
An “Apple Watch” for machines
Lim shares with DigitalEdge Singapore that the idea of sound-based predictive maintenance came from his experience as a crypto-miner. When he first co-founded his other company, Mining Rig Club, in 2017, he spent countless hours manually monitoring large crypto-mining server farms and managing unexpected server failures.
Realising that the approach was costly, inefficient and not scalable, Lim and his team decided to build a software that alerts them to a possibly faulty machine. “We wanted to develop something similar to an Apple Watch, which can diagnose the health of our servers in real-time and predict when they are going to fail. The main aim was to prevent machines from [unexpectedly] breaking down,” says Lim.
See also: From AI beds to remote ICUs, start-ups are plugging India’s health gaps
After much research, Lim learnt that sounds, which is produced by vibrations of air particles, could give a good early indication of what is wrong with a machine. He states: “Many sensors in the market, like thermal and voltage, can detect signs of failure but they don’t provide much information. For instance, a thermal sensor can show that a machine’s temperature has increased by a certain percentage but it can’t suggest what is causing the machine to heat up.”
Since every piece of equipment has a unique “sound fingerprint”, abnormal sounds can suggest possible root causes of the underlying problems with a machine. Lim, therefore, melded his keen sense of hearing — honed from his time as a principal French Horn player in the Singapore National Youth Orchestra — and interest in artificial intelligence (AI) to conceptualise Groundup.AI, which offers a sound-first predictive maintenance “operating system”.
How sound-based predictive maintenance works
See also: Connecting ecosystems: Build blocks, not billionaires
Similar to an auto mechanic diagnosing problems based on sounds coming from an engine, Groundup.AI’s solutions pick out sound anomalies that signal potential machine failure ahead of time.
Groundup.AI does so by first deploying its sound sensors on a customer’s industrial machines to gather data on their health. The data is then analysed using AI to identify the patterns and trends of the machines’ performance. Thereafter, operators can view those real-time insights through a dashboard. Since the dashboard also triggers alerts when sound anomalies are detected, operators can fix the possibly faulty machines before they cause serious issues.
How Groundup.AI solutions work to monitor machine health in real-time and provide actionable alerts for early intervention. Photo: Groundup.AI
To enhance its offerings, Groundup.AI has also developed a noise cancellation methodology. Lim explains that industrial machines are usually placed close to each other, so sound sensors tend to pick up many different signals. An algorithm that can focus on the sounds emitted from a specific device and block out other machine noises is therefore needed to get a more accurate view of the device’s health. “[To support that algorithm,] we’re building a whole library of noises from various machines — even random ones at different frequencies — so that we can block them when needed,” he states.
Since industrial companies may replace their machines or add new equipment over time, Groundup.AI’s AI system uses transfer learning, which means it can take the relevant parts of a pre-trained machine learning model and apply it to a new but similar machine or task. “We’ve trained our AI model such that it can use the knowledge it gained from monitoring previous machines and apply to a machine of a different model or an entirely new device. This shortens our AI’s learning process — even if a customer installs a new machine, they can see the return of investment of using our AI-based predictive maintenance solution within three to six months,” says Lim.
To further get customers’ buy-in, Lim shares that Groundup.AI’s solutions are designed to be easy to use. “Our software is built in a way that is zero-code and simple to understand. For instance, we’re using colours for alerts instead of having labels like ‘severe alert’ as some operators might be illiterate.”
To stay ahead of the latest tech trends, click here for DigitalEdge Section
He adds that customers can also subscribe to Groundup.AI’s managed service package. By doing so, they do not need to look at their dashboards every day but are still assured that they will receive alerts as soon as abnormal machine sounds are detected.
Accelerating the digitalisation of industrial firms
Since Groundup.AI’s solutions can help industrial firms achieve at least 95% machine uptime, over 30 multinational corporations and several small and medium-sized businesses have signed contracts with Groundup.AI. These customers are mostly from the manufacturing, oil and gas, maritime, transportation, construction and mining across the Asia Pacific region.
Not wanting to rest on its laurels, Groundup.AI is looking to increase its AI system’s ability to integrate its data with that from other sensors. “There's currently no one industrial operating system layer for machines. So our plan is to ensure that any machine data, regardless of the machine type or size, can rest and run on our platform. We want to be the standard operating system [like Android but] for industrial machines in the long run,” says Lim.
He also shares that the company will look beyond offering predictive maintenance solutions in future as its main aim is to help accelerate the digitalisation of heavy industries. “Our mission is to reinvent industrial solutions and bring innovations to industries [that used to be hesitant to use technology].”
“Predictive maintenance is a good way for industrial companies to start their digital transformation. Once they understand the transformative value of AI and IoT and see how those advanced technologies can produce results for them — such as helping them save money or even make more money — they will be more open to digitalising other processes. Groundup.AI wants to be the partner providing such solutions to support their digital transformation journeys,” he concludes.
