New ways of Managing Risk

Data-Driven Decisions are at the centre of the next generation of risk management tools. Many industries have proved the benefits of utilising data. For example, research has shown that companies implementing data-driven decisions have?output and productivity that is 5-6% higher (Brynjolfsson, Hitt, & Kim, 2011). Forrester also estimates data-driven businesses are growing at an average of more than 30% annually (Forrester, 2018).

In a data-driven world, the risk is a function of the ever-changing operational status of an asset (e.g. process plant or ship). Data emerging from the constant change in operations can impact both the consequence and probability side of the risk equation. So, real-time information, as well as historical data, plays a central role in supporting new risk models.

The following section describes the application of data-driven solutions in the context of risk management.

Predictive Maintenance and Condition-Based Models

Predictive maintenance is an example of a new technique for improving risk management. The installation of smart sensors to machinery gathers data related to specific parameters of an asset. These parameters range from vibration, power consumption, temperature and more. Data scientists create models with this information to identify the normal behaviour of assets and when something abnormal happens (e.g. an emerging failure or a disturbance to the process).

Figure: Anomaly Detection with Predictive capability @Falkonry

The definition of reliability?(Naresky, 1970) offers another insight into the importance of this new risk technique:

“Reliability is the probability that an item will perform its intended function for a specified interval under stated conditions”.

The part mentioning "under stated conditions" infers that each equipment has its own useful life, based upon the condition that each equipment is operating. So, two pumps of the same type (e.g. centrifugal pump), in different operating scenarios such as flowing water and oil will have different "reliabilities" over their lives. Now, it would be simplistic to assume that companies should break the analysis down into two distinct equipment classes and move on. The main reason this would not work is that each pump will then have its operational condition such as pressure and temperature, which may also vary. Furthermore, the number of assets of similar classes in complex plants tend to make this solution unmanageable. Predictive Maintenance helps companies to solve this problem by collecting data from individual equipment and identifying their specific behaviour over time. The result is a prediction model which silently acquires data and, when something out of the ordinary happens, it raises an alert so that maintainers can take action. ?

Once data from multiple assets are available, the data scientist can start to explore whether there is any correlation across equipment of the same and different types. Several techniques enable data scientists to cluster information from various assets and show patterns that are leading to anomalies. This approach elevates predictive maintenance beyond the equipment specific prediction models, which are generally based upon particular variables concerning the equipment operation. It allows for benchmarking of performance and key performance indicators.

Figure: Historical Analysis, tracking abnormal events @Falkonry

Finally, modern technologies focus on offering the ability to consume data in real-time and explore vast amounts of historical data which are usually stored in data lakes. So, as new data becomes available, new models and correlations can be identified, creating even more powerful predictions and classification of events. The evolving nature of these models means that the technology cannot be a limiting factor. So, a lot of technology companies are investing in supporting big data tools that are functional regardless of the infrastructure and, sometimes, underlining software.

Four major technology trends have been central to empowering companies to benefit from predictive maintenance. The Industrial Internet of Things (IIoT) combined with advanced Big Data Analytics, powered by cloud computing technologies and advancement in cyber-security. The next chapter of the whitepaper investigates these trends.

Major Technology Trends

Four major technology trends have been guiding the deployment of industrial techniques:

• Industrial Internet of Things (IIoT): IIoT is a network of sensors which are typically connected to assets to monitor and collect data. This data can be further analysed using big data analytics techniques.
• Big Data Analytics: the following 4Vs defines Big Data – veracity, velocity, volume and variety. Being able to process vast amounts of data (volume) from multiple sources (variety) with significant ingestion rate (velocity) still confirming the integrity of data (veracity) is a great challenge. The technology industry has been using specific solutions to treat, process and visualise. Data science is the scientific field utilising this technology to create data-driven solutions. Data science is a multi-disciplinary field that uses scientific methods, processes, algorithms and systems to extract insights from structural and unstructured data. Data science is related to data mining and engineering. More recently, the industry has turned its attention to developing a range of tools automating data science. These new tools are aiming at speeding up the model creation process.
• Cloud computing: cloud computing is the delivery of computing services – including servers, storage, databases, networking, software, analytics and intelligence
• Cyber-security: cyber-security is the practice of defending hardware and data from malicious attacks.?With companies relying more and more on data,

To support the implementation of these new technologies, companies have been investing in developing Digital Industrial Platforms. The European Union defines industrial digital platform (European Union, 2018):

Digital industrial platforms are essential, for instance, in connected smart factories, where the platform can take data from the machines, make it accessible to monitoring and control applications. This technology stack also allows third parties to develop applications based on that data and connect different users and application developers.

This definition uses the best parts of the major technology trends to support companies in making informed decisions about maintenance and operations.?

One key enabler for the increasing number of industrial digital platforms and, in general, software solutions is the advent of Open Source projects. Open-source software (OpenSource, 2020) is designed collaboratively by members of a specific project and, more importantly, it is distributed under an open-source license. This license agreement grants all the rights to use, study, change and share a modified or unmodified for of the software.

The last part of the technology equation is an online marketplace. A marketplace can be used to expose and orchestrate digital content, fostering innovation and collaboration.

All these technology trends have been changing the landscape of industries all over the world. However, the industry has underestimated the implementation of these tools. The next section explores the implementation pitfalls.

Implementation Pitfalls

Starting with the foundation of any analytics solution: Big Data and Data Science. In 2016, Gartner estimated that 6 out of 10 data science projects were failing. In 2017, one of Gartner analysts said it was likely closer to 85% (TechRepublic, 2017).?Interestingly, the leading cause for a project failing did not involve the technicalities of implementing tools. The most significant barrier was "insufficient organisational alignment, lack of middle management adoption and understanding and business resistance”. In addition to this crucial challenge, a few other points have been raised:

• Solution not solving any specific problem because of lack of end-user engagement
• Inability to ask practical questions which leads to re-work, making projects extremely expensive
• Lack of a prototyping phase has also been listed as the cause of projects failing. Companies go directly to the development phase without validating fundamental scenarios and applications

Other sources point out the lack of Subject Matter Experts as a central piece in a data science project (Forbes, 2019). Without expert feedback, projects are likely to under-deliver value to companies.

The movement called Industry 4.0 (BCG, 2020), or the Industrial Internet of Things, has not expanded as expected either. In 2011, Cisco predicted 50 billion devices (Evans, 2011) would be connected by 2020 – this number accounted for both the consumer and industry market.?The research focused on industrial devices, in 2017, Gartner (2017) forecasted 8.4 billion devices connected. Recently, Gartner (Gartner, 2019)?has estimated only 5.81 billion devices, 30% fewer devices than initially expected. This finding aligns with data coming from a survey (Eseye, 2019) where companies have suggested they underestimated the complexity involved in an IoT project 6 out of 10 times. The main reasons for the disappointing growth of the Industrial Internet of Things include integration problems caused by a fragmented market to the lack of skills in communication protocols made IoT projects complex and costlier.?

From the four major technology trends, cloud technology is the one that seems to be attaining the best score regarding its dissemination.?Even though cloud technologies is faring better than other major trends, its adoption is still limited with around 20% of organisations adopting public or private cloud technologies, according to research (McKinsey, 2018). One of the leading causes used to explain the low adoption rate is that many applications are still using old architectures and the cost to migrate these old solutions is high.

The next article explores two use cases which are offering actionable insight.?