BLIND FAILURE PREDICTION
My friend Stephen Allott (we worked together at Micromuse) sent me a few weeks ago the post he wrote about the benefits of the RoCE model (Return on Capital Employed). Originally designed by McKinsey some years ago, He used it to help startups’ customers to calculate the return on the capital they invest in technologies.
The purpose of this model is to formalize and quantify revenue gains and cost savings foreseen by the implementation of new technologies.
I thought it would be interesting to apply this approach to our domain: failure prediction of IIoT-enabled equipment. Especially, nowadays, when budgets are challenged and under tight review.
The RoCE model
Let’s start with the standard model and see how it works.
The idea is to split gains and savings into four categories: revenue, costs, working capital and fixed capital.
Based on past projects developed with our customers, we illustrated each branch of the RoCE tree with the benefits actually obtained and most of the time quantified. We identified nine of these in total.
The objective with this document is to help and guide companies to evaluate RoCE related to the acquisition of Amiral Technologies’ Failure Prediction Solution “DiagFit”.
To build their business case, we assist each company to quantify gains and savings. Each branch might be quantified specifically and sometimes only one or a couple of branches are enough to reveal tangible gains, depending on the customer’s use case.
In the generic analysis below, we provided estimated numbers wherever that was possible.
1. More customers: win more customers than competitors
When a company adds failure prediction features to his manufactured equipment, associated benefits contribute to differentiating his offer from his competitors’.
- Contractually increased equipment uptime
- New way to support his new Opex-based sales model (selling hours of usage) versus the traditional Capex-based sales model (selling equipment),
- Featuring an “AI-powered” label on their product with positive impact on brand equity,
- New features such as “wear and time before failure indicators” on some critical components of their product.
2. Provide new services to customers
As failure prediction helps anticipating and correcting failures before equipment downtime, equipment manufacturers can propose SLA (Service Level Agreement) with higher uptime and quicker recovery, with the option to accept higher penalties if needed to secure new deals. These agreements represent significant upsell value compared to existing ones.
Features such as providing a car part wear indicator lead to increasing the revenue per service or per product. Car manufacturers can price premium options coming with wear indicators displayed on their digital UI, for components like tires, brakes, …
3. Faster and more revenue per service or product
On the issue of time to market, as one of DiagFit’s major innovations is the possibility to start predicting failures without needing historical failure data, launching a failure prediction service can be put in place in a few weeks compared to months/years for historical data collection.
In terms of revenue, an equipment sale can start generating revenue from premium SLA as soon as it is sold, in the car industry, subscription to these new services can lead to additional revenue in the range of €5 to €20/month/car.
4. Maintenance costs reduced
As the equipment uptime increases, costs of maintenance are reduced in four ways:
- Less intervention time and travel for maintenance experts
- Better scheduling of maintenance operations in a preventative manner instead of a corrective manner, thus saving the costs of equipment downtime.
- Anticipation of ordering the required spare parts, better management of parts inventory and supply chain (this is significant for industries such as aeronautics where the spare parts need to be dispatched all over the world)
- Less penalties related to SLAs.
5. Activity needs fewer heads
To adapt to fast changing markets especially in energy and transportation, industrial companies must move from a volume-based business model to a value-based business model by incorporating digital technologies in their products and services. Attracting talents mastering Machine Learning technologies is becoming an increasing challenge for them versus banks, insurances, retail and digital companies.
With DiagFit’s intelligent automation, the software is accessible for use directly by maintenance operators without requiring data scientists, or experts with programming or data analytics skills.
As DiagFit models are mainly data-driven, domain experts will contribute to the software set-up and for providing feedback on the generated alerts, however, the system does not need a domain expertise to derive physical models.
In addition, less maintenance operators are needed at facilities to monitor equipment.
6. No need for expensive cloud resources or GPU-based servers
One of the major innovations incorporated in DiagFit is its low computational appetite. The technology powering DiagFit allows running the software on common hardware, either in the Cloud, using low resources Edge computing or embedded in the equipment.
A study showed that the end-of life diagnosis for a one-sensor equipment can be embedded over a low-grade built-in processor.
TCO (Total Cost of Operation) is then highly controlled and minimized versus traditional AI-based software requiring resource intensive capacities and GPU-based servers to run them.
Savings can be estimated in hundreds of thousand Euros per year.
7. Pre-production costs lowered
As DiagFit can start predicting failures without needing historical failure data, there is no need to collect and store massive historical data over several months or years in order to collect representative failure data to build the predictive model. DiagFit is mostly used with no or very few historical data.
Savings are calculated by:
- minimizing time and effort to collect historical data
- minimizing storage capacity costs