Reducing downtime and improving efficiency with AI

Leon Lim, CEO & Founder,

Companies worldwide face increasingly complex challenges, competition, and potential revenue loss as a result of unexpected machine failure. In fact, the world’s largest manufacturers are losing close to $1 trillion a year due to machine failure. This trend has been further exacerbated by the COVID-19 outbreak, which has exposed the weaknesses of business models in the industrial sector.

For instance, the longstanding reliance on manual labour took a toll on industrial productivity when social distancing and working from home became the norm. In spite of the surge in AI adoption arising from the Fourth Industrial Revolution, this trend is only largely prevalent within telecom, high-tech, and financial services sectors, with other industries like manufacturing and pharma industries trailing behind.

I would like to think that industrial companies in the manufacturing, maritime, construction, oil and gas, and transportation sectors are the backbone of our economy, and a lot more can be done to help these big players smoothen their transition into the Fourth Industrial Revolution.

The problems of unplanned downtime

As infallible as they are, heavy machinery is not immune to errors. In fact, even the slightest malfunction can lead to sudden downtime. I have spoken to enough companies to know for a fact that unplanned downtime is a major problem that transcends all industrial companies. No one is spared from this because most, if not all, rely heavily on machinery for their day-to-day operations.

As such, not only do machine breakdowns hinder productivity, it also brings about operational nightmares and incurs gargantuan costs for companies handling the sudden catastrophe. Not to mention, sudden breakdowns of massive industrial machinery can also pose great danger, threatening the safety of workers on site.

The current approaches to industrial asset monitoring using AI are expensive and often require large teams of skilled personnel to oversee operations. For some companies, that is a tall order. In fact, over 40% of companies cited skill shortages as one of the biggest barriers to their adoption of AI. To add insult to injury, the inability of existing legacy systems to interface with AI is also a huge dealbreaker.

On the ground, companies also raised concerns about AI replacing human workers, and it dawned on me that there was a lack of understanding about how AI can be used more effectively in their industries to maximise growth. Leveraging AI solutions as critical maintenance and monitoring tools can help industrial companies mitigate equipment failure in advance before catastrophic failure occurs.

As a result, companies can produce their products more reliably and cost-effectively, possibly even freeing up time for workers to upskill themselves and take on value-added responsibilities on the job.

A sound-first predictive maintenance solution to tackle unplanned downtime

Just as the rattling engine as one cranks the ignition of a car indicates an imminent car breakdown, machines often give off different sounds when there are underlying problems. For sound-based predictive maintenance solutions, these differences in sounds can easily be picked up by the sound sensors that emulate the diagnostic hearing abilities of experienced workers, and be flagged out to engineers pre-emptively before it escalates.

A sound-first predictive maintenance solution consists primarily of two steps – Condition-based Monitoring and Predictive Maintenance. Condition-based Monitoring offers real-time equipment diagnosis akin to monitoring the health of machines in real-time, like an “Apple Watch” for machines, and Predictive Maintenance functions like a “crystal ball” to help companies predict equipment failures beforehand.

To effectively diagnose the health of machines, AI solutions need data to refer to. AIs are not immune to making mistakes and machine learning takes time to perfect and become useful. To expedite the process, a good AI-based machine maintenance solution should have an existing data reservoir that existing legacy systems can use as a reference to detect anomalies, without the need for companies to start new training models from scratch.

What’s more, a holistic sound error database can further help companies to get a head start in deploying the solution and identifying machine faults with little time needed for calibration, which eases retrofitting for them. Over time, self-learning AI technologies also do become more accurate as more data is collected.


In my opinion, there is no doubt that predictive maintenance solutions will become an indispensable tool in the industrial sector in the coming years. According to a 2019 report by Allied Market Research, the global predictive maintenance market size was initially valued at $4.3 million in 2019 and is projected to increase more than 7 times to $31.9 million by 2027. With the help of machine learning technology, I am confident that sound-based predictive maintenance solutions will become an OS layer for all industrial machines – just like what Microsoft did for PCs.