Sign up for our newsletter and get the latest big data news and analysis.

In the Time of Big Data and Machine Learning, It’s Important to Ask “Why?”

In this special guest feature, Sundeep Sanghavi, Co-founder and CEO of DataRPM, believes that “The Five Whys” interrogation technique developed by Sakichi Toyoda, founder of the Toyota car company and hero of the Japanese industrial revolution, is still very relevant in today’s age of big data and machine learning. CEO Sundeep Sanghavi co-founded DataRPM, a Progress company, with the goal of providing a cognitive platform for predictive maintenance solutions to organizations challenged by the volume, velocity and variety of their big data and machine learning

Imagine a robot stops working on a factory floor and engineers are trying to fix it. The scenario follows as such:

Engineer 1: “Why did the robot stop?”

Engineer 2: “The circuit overloaded and blew a fuse.”

Engineer 1: “Why is the circuit overloaded?”

Engineer 2: “The bearings were insufficiently lubricated.”

Engineer 1: “Why was there insufficient lubrication?”

Engineer 2: “The robot’s oil pump is not circulating enough oil.”

Engineer 1: “Why is that?”

Engineer 2: “Because the pump intake is clogged with metal shavings.”

Engineer 1: “And why is it clogged?”

Engineer 2: “There is no filter pump.”

This scenario is an example of “The Five Whys” in action. If you’re not familiar, the Five Whys was developed by Sakichi Toyoda, founder of the Toyota car company and hero of the Japanese industrial revolution. The Five Whys is an interrogation technique, formalized in the 1950’s, designed to explore cause-and-effect relationships as related to specific problems.

The number of “whys” is taken from the anecdotal observation that – more often than not – five iterations of “why” are needed to find the root cause of any given problem. The thinking behind this being that asking “why” of the situation five times would be sufficient to find a way to solve problems with machinery and processes.

The five questions presented in the above scenario, asked in succession, have identified the root cause of the problem, which now puts engineers in a position to fix it. Originally intended for an earlier industrial age, the technique of the Five Whys is as relevant as it has ever been. Because, when integrated with today’s modern technology, such as machine learning, data science and cognitive computing, the Five Whys are propelled to a new level in which they do not simply identify the root cause of a problem, but also provide prescriptive measures to avoid the failure.

Why is The Five Whys Still Pertinent?

One big reason why the Five Whys remain crucial for modern industries is recalls. They remain a constant and ever-present problem within manufacturing, resulting in high costs and – at times – shattered customer confidence for safety.

The reality of connected factory is rather complicated – machines work 24/7 and are subject to rigorous wear and tear daily, which can cause unexpected equipment failures. Data is the fuel for the Industry 4.0 engine and accurate data analytics is the prerequisite for successful implementation of predictive maintenance based applications.

This is because, even though modern factory floors and are filled with sensors plugged into the Industrial Internet of Things (IIoT), the ability to sift through and identify the right signal amidst the noise is tough, time consuming and complex.

While typically what is used are superficial, standard, formatted monitoring tools and dashboards, what happens beyond these beautiful charts and can companies get to where the real answers lie? It is important for companies to move from a mere phase of discovering facts which is still reactive and preventive in nature. This could be gravely misleading as it provides only partial indicators of when the failure happens (rather than digging into the root cause of understanding what caused the failure or how it can be avoided completely).

Looking For the Needle? Make Those IIoT-Data Haystacks Smaller

The future of the IIoT involves billions of connected ‘things’ which will generate trillions of gigabytes of data – all of which result in creating a market of trillions of dollars. At the same time, the IIoT has resulted in creating a massive upswing in data volumes due to the increased refined granularity of the data being produced. Today, data science and machine learning are dealing with very specific challenges with this volume of data that is being produced. Finding anomalies itself can be a daunting task here as most of these are extremely rare events, and hence analyzing these sporadic events itself is quite difficult. Now within this context, anomaly detection of these rare events in such humongous data sets by conducting manual in-depth analysis and visualization of very large data volumes can be equivalent to looking for a needle in a massive haystack!

Further, deeper questions with regard to the data itself emerge: checking if the data is relevant, if it is of sound quality, and if the model is produced by an algorithm translating from a mathematical relationship to a causal one. All of these need to be answered.

While this integration of the human element and technology is available today, sadly, this has yet to be properly implemented. For example: Takata, an airbag manufacturer, has been caught up in a high-profile recall sparked by customer injuries and even deaths. The recall became so widespread that the company filed for bankruptcy in June, while many injured drivers are filing legal action.

Had Takata implemented machine learning and predictive maintenance technologies that employed The Five Whys, could this unfortunate series of events been avoided? We will never know.

The Catch

There are, however, some limitations to the integration of the Five Whys and modern technology; no system is perfect. In using this technique, it can at times be difficult to distinguish between causal factors and root causes. Sometimes there is also a lack of rigor in deducting when advocates of the system are not required – and subsequently do not – test for the sufficiency of root causes generated by this method of industrial soul searching.

The trick to solving this stumbling block, and ensuring the Five Whys remains relevant, is to keep it grounded in observation. This, combined with skillful use of the system, will keep the method a viable one to bolster Big Data and machine learning technology now and in the future. To capture the significant gains from machine learning and the industrial IoT, companies will have to automate underlying processes that will help them scale with ease for ‘insights’.

The Future Marches On

While sensors can help humans interact with machines, today, cognitive predictive maintenance technologies will over time evolve and become “smarter” and more intuitive. As this happens, we will essentially automate the Five Whys technique. New machines with greater ‘thinking’ capabilities can employ the interrogation technique across thousands of factories and millions of sensors. This level of scale, of course, is not possible with a human workforce.

Machines can also come up with an answer that requires brief and non-technical approval from a human supervisor, so that maintenance teams know what questions are being asked to reach the appropriate conclusion. For example, General Electric has hundreds of factories across the globe, and there’s just no feasible way that it could have human teams ask all the ‘whys’ needed to avoid and prevent technical failures. With the advancements in predictive analytics and cognitive machine learning, machines can do that for GE, providing fast and actionable solutions in the process.

As sensor use continues to grow and proliferate, the benefits of machines using the Five Whys will be felt across the industry. Sensors and factory equipment will combine over the IIoT, making advanced warnings and predictive maintenance better and more advanced. This will in turn result in safer factories, fewer recalls and more protected workforces for organizations.

Here’s to asking why.


Sign up for the free insideBIGDATA newsletter.

Leave a Comment


Resource Links: