The world of industrial analytics is becoming increasingly crowded with offerings from data science houses (consultancy) and data science programming tools (DIY), making it easy to become confused.
How do these offerings compare? Do they speed up the time to implement predictive maintenance and process optimisation? What are the advantages and disadvantages?
DATA SCIENCE HOUSE
Think of these companies are like an off-site data science team – with a price tag! Yes, they build bespoke models of your assets, however there are a few things to watch out for;
- This is still a manual data science solution – often taking just as long to deliver results as an ground-up in-house build would take (think 3-6 months)
- Initially models are built with limited historic data and then moved into a live production environment, following which any changes and management of the models will have to be done by the Data Science House (if at all)
- Models are often built using manual AI mathematical modelling techniques – however with limited historic data
- Models are transferred from client to client and between equipment, whereas most operators will tell you that two identical pieces of equipment rarely perform identically
- Customers only choose to model specific critical pieces of machinery due to the costs
- ?Data science houses typically provide prototyping solutions only, making it difficult to rollout predictive maintenance regimes client side
DATA SCIENCE TOOLS
With a great variety of tools on the market, some built for industry and some generic, these solutions empower your existing data scientist to model problems themselves. A couple of things worth noting;
- The data scientist will still need to engage in the more manual time-consuming preparation tasks, which include loading, querying, parsing, filtering and sorting of data
- The typical result from the modelling is condition monitoring, providing you alerts based on the current condition of your assets, rather than the future condition
- Most Data Science tools rely on a library of pre-created models, which can be AI mathematical models or physics based models based on original manufacturers’ (OEM) design specifications. This limits the number of model queries that can be run by the data scientist using the tool
- Some data science tools require the customer to use their proprietary IoT sensors
- Programming knowledge is required for the data scientist to build and deploy the model
- Analysis is generally limited to individual equipment based on highest priority business needs, and modelling does not look at the data from the whole facility
- Once the business is confident in the model, it is exported and put into the live production environment, at which point it is maintained by IT and the data scientist is no longer involved in maintaining, validating, debugging or retraining.
- The typical manual data science process is still followed, regardless of using the Tool. Individual problem solving can still typically take between 1-3 months
At VROC we saw an opportunity to automate the above processes, providing AI predictive analytics for ‘whole of facility’ predictive maintenance and optimisation. This has been achieved through the development of an AI platform which doesn’t rely on pre-built models, nor programming skills and empowers subject matter experts and engineers by automating data science process.
Learn more about the differences between a traditional data science approach and an automated data science approach in our whitepaper ‘transitioning from traditional data science to automated data science’ – download it here
