Cost-effective use of relevant data - the InsightProducts case

With sensors becoming rapidly cheaper, manufacturing companies are expecting their massive sensor deployment to support vital services, such as predictive maintenance, automated supply chain management, operations & usage monitoring. However, billions of connected IoT devices, generating monumental amounts of data do not necessary result in smart systems or products. They do have a direct impact on cost. The InsightProducts project supports companies to improve their product and service offering through a cost-effective approach to relevant and qualitative product data acquisition and use in view of digital servitisation.

It is a 'Brave New World' for product manufacturers, with sensors, wireless communication technologies, software and XaaS models becoming more prevalent, products are ever smarter and more connected. Data-driven product management is vital to decision making in view of deepening the relation with customers in real time. Therefore, big data analytics is still seen as the key to ROI for product manufacturers.

However, the current big data approach is often seen as a procedure with two basic steps:

  • Step 1. Instrument the product or system with an abundance of sensors generating large amounts of data.
  • Step 2. Drop the resulting data lake to a data scientist and expect 'something' to happen.

The problem with this approach is that contrary to what one might expect, less than 1% of these big data is actually analysedThis is inherently linked to the costs made for acquiring these large data pools. Although sensors are nowadays becoming much cheaper, the complexity of the supporting infrastructure and amount of generated data that comes with them is not a viable solution any longer. Many hidden big data costs are often overlooked, such as:

  • Sensor & network management - monitoring, maintenance, repair, QoS guarantees for data communication to backend.
  • Costs of data quantity vs. data quality - backups of huge amounts of data, storage & processing of low quality data, redundancy, manual effort of data transformation & integration, manual effort of annotating/labelling data, and at the end of the line: manual effort in extracting meaningful and actionable insights (beyond rudimentary statistics) due to lack of domain knowledge of the data scientist.
  • Security & privacy concerns - often third class citizens, but key investment in these covers all the layers, from sensor, communication, storage, to analytics.

Of course, if you are lucky (or unlucky depending on how one sees it) enough, lack of access to customer data to achieve something intelligible with it, might have cost you the setup of an entire backend infrastructure without actual data to work with.

Cost-effective use of relevant data

Hence the premise of the InsightProducts project, led by Sirris, focusing on decreasing obvious and hidden costs of big data and targeting qualitative data instead by:

Optimising sensor deployment and decreasing the amount of data sent

The process of defining optimal InsightProducts architectural layers involves splitting up the problem in the following decision points:

  • Selection of suited communication & sensor solutions - depending on several application requirements, such as operating environment, usage & fault-tolerance resulting in optimal communication & messaging in view of relevant data acquisition.
  • Optimal use & placement of sensors - in view of qualitative data acquisition, the amount and location of the sensors plays a major role.
  • Qualitative data - as result of preprocessing, annotation & removal of redundant measurements.
  • Data & system security - including communication, data and physical security of sensors.

Improving data quality and providing actionable insights

Providing actionable insight into products is key for a company to ensure a qualitative product tailored to the different usages by its customer. These insights help the company in understanding the product usage better and can pinpoint the weaknesses and strengths of their product. To define actionable insights, the product needs to be monitored correctly and the right data needs to be collected. There are several steps that a company can take to acquire qualitative data. Some of those are guidelines to consider before gathering data, while others can improve the quality of the data after acquiring it. A qualitative dataset makes it easy to enrich the data using third-party datasets, gives the data scientist the right information to extract the right insights and provides the company a structured format to capture data in the future.

On 22 October 2019 Sirris organised its first annual workshop open to the public detailing on the research topics of the InsightProducts project, as mentioned previously. During this workshop, two themed sessions on customer data access and digital servitisation were organised consisting of company testimonials from Gilbos, 3E, Renson, Certis, and be.wan highlighting obstacles and approaches.

Sirris was also present at the Flanders Make Symposium on 26 November 2019, where we presented a preliminary set of results on the five identified demonstrators of the InsightProducts project. Three of these demonstrators are technology-oriented and aim to illustrate different aspects of the process of acquiring and exploiting qualitative data:

  • Optimal use of sensor & communication solutions supported by and supporting qualitative data.
  • Determining how to acquire qualitative data and extracting actionable product insights from it.
  • Enhance design and operations of a product, based on data insights.

The remaining demonstrators are methodology-oriented and aim to provide companies with success stories, best practices, etc. related to:

  • Digital servitisation: supplement traditional product offerings with services and solutions.
  • Gaining access to customer data.

Follow-up initiative

In addition, as a result of the interest on the current demonstrators and of the extensive follow-up discussions with companies, a topic for a follow-up initiative has been identified, focusing on trust and transparency in data analytics. This entails:

  • Dependable and trusted data sources - Ensuring data used in analytics are real, accurate and have not been manipulated
  • Reliability and trust in data analytics processes - Ensuring data analytics is performed transparently and the extracted knowledge is reliable and trustworthy.

Are you interested in the knowhow of the InsightProducts project or the follow-up initiative? It is still possible to join the project’s user group, so you'd have a front row seat to the project demonstrators outlined above and benefit from the knowledge developed within the project. You can find more information on InsightProducts here.

With the support of: