Sooner or later the data science jargon and marketing hype is going to subside, and manufacturing companies, among many other sectors, are going to find themselves sitting with broken promises. It is therefore important that these organizations understand clearly how they stand to benefit from and be empowered by data science and the challenges thereof. And with this in mind, In this article, I discuss the opportunities, challenges and potential sources of data associated with data science for manufacturing companies.
If there is one sector that is set to benefit immensely from Big Data, it is manufacturing. Every individual and company are influenced by manufacturing one way or another, and the industry is sitting on vast amounts of data. And with an ever increasing appetite for affordable high-quality goods due to a rise in living standards across the world, their only hope to satisfy this need lies in Big Data analytics. If manufacturers manage to harness this power of data they would be able to, for example, predict with near certainty consumer needs and manufacture a product that will be widely accepted.
Meanwhile, It's interesting how there is often an effort to distinguish between consumer and Industrial IoT because when it comes to data science for manufacturing a real opportunity to improve efficiency and quality lies in combining the two. Whereby data is gathered from consumer goods under real world conditions and supplied back to manufacturing systems for analysis. Putting a wave of entirely new revenue streams well within reach.
But that is just part of the story. What holds my interest and what is perhaps the lowest hanging fruit as far as data science endeavors in manufacturing go is the possibility of automating the analysis of data from equipment sensors to detect anomalies and predict equipment failure. Equipment lifetime and uptime can also be predicted, defects identified and optimal scheduling of inspection rounds achieved. This presents an unprecedented opportunity to reduce downtime and increase machine utilization. As they say, a manufacturer is only as good as the machines that produce its products.
Here is the kicker though. In a bid to guarantee a particular product lifetime, manufacturers tend to design a product to be more robust and complicated than necessary for its application. This, in many cases, significantly increases the cost of production and consequently that of goods. However, if sufficient data is collected and analyzed while the product is in usage, it is possible that elements which do not influence product lifetime can be identified and lessened, resulting in huge savings.
And it doesn't stop there; the real game changer is the application of advanced machine learning to enable manufacturers to model products, machines, and assets in software to simulate different scenarios to find ways to maximise efficiency for any given situation.Providing them with priceless insight into the design, usability, and serviceability of a product before resources are committed to its production.
These are just a few of the many opportunities data science presents to manufacturing, including purchase order automation based on operations data and opening new revenue streams by leveraging data to deliver exclusive experiences for which customers can pay more.
On the other hand. It's not all sunshine and roses as the industry still faces some challenges. Building a business case for data science projects is of paramount importance, yet manufacturing use cases are not as prominent as those of natively digital companies. This presents a challenge because businesses that fail to build a case will certainly not get a return on their data science investment. The situation is made more challenging by the fact that data in manufacturing plants is not readily available in digital form. Instead, unstructured and "dark data" is more prevalent.
What's more, the implementation of Data Science endeavors in manufacturing faces a technical challenge.This is because distributed computing frameworks such as Hadoop which are primarily used for Big Data across different sectors have their origins in search engine technology. Yes, they might have evolved to become applicable to a variety of industries, but they were never built with Manufacturing in mind. I stand to be corrected, but I am convinced there is no big data framework distribution specifically tailored for manufacturing scenarios. The point being that making current structures work for manufacturing would require some form of adaptation.
As if that's not enough, data science happens to be a discipline that is not fully replicable. It requires constant evaluation as the data changes. That may be of little concern to big companies like Amazon or Google that can afford to put together a team of data scientists to consistently attend to the changes. But for a small or medium sized manufacturing company that has to deal with a flood of dynamic and multivariate plant data, to have to an in-house team would be costly but somewhat necessary.
As it turns out, to keep up with the ever increasing customer expectations, manufacturers must be prepared to build a single view of their customers into the entire process. Approaching a batch size of one. The challenge lies in collecting IoT sensor data and harmonizing it with external sources of data.
Data can be obtained from a lot of sources in manufacturing, but first things first. Manufacturers must begin by laying the ground work for a secure environment before they can start pulling precious information from assets and machines. Simply put, consumers are already providing valuable data to manufacturers unbeknown to them in most instances. For example, most modern vehicles, mobile phones, and machines e.t.c have an inbuilt capability to track location and performance.
Every touch point along a value chain also presents an opportunity to collect and carefully document data. From purchasing, supply, sales to maintenance and so forth. Data can also be collected and analyzed from social media and visitor actions on the company's website in order optimise and individualize consumer interaction and conversion.
Even better is the fact that internal systems of data collection and analysis such as plant historians and ERP systems are a treasure trove of data. They can provide operational, process and product data. Classic methods of data collection like customer surveys, focus groups, and call centers cannot be overlooked as they ingest much needed unbiased information. I say this because abstract data is easy to become a victim of self-fulfilling prophecy.
Machine to Machine interactions and the use of IoT to collect data from assets, machines and products machine readings can perhaps be the largest source of data in manufacturing scenarios. Service manuals and maintenance log sheets referred to as dark data are also a source of valuable data even, one that requires manual mapping into digital systems. Advancements in natural language processing are however being made to automate the capture of dark data, a move that would significantly support data science efforts.
Fun fact: Veritas Global Databerg report found that 85% of stored data is either dark or redundant, obsolete, or trivial.
In conclusion, the importance of good quality data cannot be overemphasized. Poorly collected data is of no value to the business as it will affect interpretation. Data interpretation is after all the epitome of its value to business. So manufacturers need to be cautious about collecting data just for data's sake.