AkzoNobel, a leading supplier of paints and coatings, is pushing into data driven services which help marine customers plan maintenance and reduce fuel costs
Some years ago, AkzoNobel decided it could not rest on its laurels and needed to embrace the data revolution. But how does a company that sells physical products like coatings, for physical things like ships, become data driven?
A team of data scientists, technical experts and marketing managers came together to find out. An early success was Intertrac Vision – developed with Tessella – which predicted fuel and CO2 savings from different coatings for ship hulls.
In shipping, tiny efficiency improvements can save fortunes in fuel costs. Intertrac Vision gave AkzoNobel a differentiator, allowing its customers to make data-driven decisions on optimal coating selections. Intertrac Vision consultations have resulted in tangible additional coating sales for AkzoNobel.
But AkzoNobel’s Marine business wanted to take things further. They recognised that the data they had acquired, and models they had developed, had huge value for the shipping industry. They embarked upon developing new data products and services which they could sell to their coatings customers and others.
Location of the deep sea fleet shown with fouling risk factors. This complex calculation, and others, need to be performed at scale for the history of the fleet to offer insights on fouling, corrosion and vessel performance.
The challenge: Building new corrosion and fouling prediction models with big data on legacy systems
When AkzoNobel embarked on the initial project, they were taking a bold leap into a new area. Like any experiment, they couldn’t have foreseen what it would become. The Intertrac Vision project used a large dataset of vessel locations, based on the AIS anti-collision system. They combined this with their historical vessel inspections - collected over 40-years - to model fouling (the accumulation of unwanted materials on ship hulls) and the effects of fouling to predict the effect on the hull performance of vessels.
Vessel hulls need to be treated and repaired in dry docks, where they are inspected so work can be performed on their hull. Booking dry dock time is expensive, so there is huge value in being able to predict what needs doing: too little time to make repairs means the ship won’t be ready in time; too much means money and time wasted. With the proven success of Intertrac Vision in the marketplace, AkzoNobel realised they had an opportunity to further exploit their unique expertise and data sets to give new insights to vessel owners.
AkzoNobel worked with their research partners to develop new models based on the corrosion breakdown of the underwater hulls of vessels. They also developed algorithms that allowed them to identify dry dock dates, and further refine the precision of their fouling insights.
However, as ambition grew, the underlying infrastructure on which the models ran became too slow for increasing demands, and too limited for new ideas. For example, pulling out data on where vessels had travelled over the past year took 60 seconds; running an algorithm on all ships in their dataset took three months. This was too long to create the rapid insights they wanted to sell. And the two-terabyte storage was full.
The solution: Data infrastructure fit for a data-driven organisation
Tessella worked with AkzoNobel to develop a new architecture for storing and managing data, which would allow them to run existing models faster, and build new models more easily.
The large datasets from shipping routes were initially fed into a cloud storage solution. These went through a data cleaning and enrichment process, which included assigning a fouling or corrosion risk score based on the locations they had travelled through. This processed data is stored separately, allowing all the operations to be easily re-run on the same dataset as the algorithms are refined.
Tessella then implemented the models on the new system, where they could run quickly on clearly curated, processed datasets. The results are calculated nightly, and aggregated together in a structured database. This allows the results to be retrieved, sorted and filtered extremely quickly and is used to provide new insights on vessel performance.
Perhaps most significantly, the new architecture allows models to run at scale. By allowing hundreds of simultaneous model runs, Tessella have drastically reduced the calculation time for the whole data set from months to hours. The new model implementations were also designed to run incrementally. This means that AkzoNobel can process new data (eg the past 24 hours shipping data) to modify existing outcomes without needing to re-process the entire dataset each night.
Michael Hindmarsh, Incubator Lead at AkzoNobel said “Tessella brought a unique combination of understanding of the business challenge, the science behind our models, and the technology options. This enabled them to quickly identify and create the right solutions”.
Resulting aggregate fouling exposures for each vessel in the deep sea fleet after processing by our new parallel calculation architecture.
Impact: New services from AkzoNobel
The project reduced model runtimes from three months to eight hours, allowing AkzoNobel to provide a near real time corrosion prediction service for entire fleets of ships.
As a result, this work has led to the development of new MVP’s such as DryDoq Insights: a tool which is developed by Tessella and currently being trialled which aims to utilise the new data package to predict corrosion and fouling, helping vessel owners make more informed dry-docking decisions.
Meanwhile, switching the existing tooling to the new data package is estimated to save eight days of sales time per month, thanks to increased data retrieval and processing speeds. Tessella is now helping AkzoNobel explore the ability to sell access to the underlying APIs to 3rd parties, allowing it to be built into industry standard platforms, vessel owners dashboards, or used in new ways that we have not yet thought of!
As AkzoNobel moves forward with developing new innovative services, they have a system which allows new models to be more easily developed and scaled up as they move from prototype models to customer ready predictive services.