DINGO’s Chief Operations Officer, Gary Fouche, thinks data is one of the most important drivers of the business. We asked him how data is made even more valuable when predictive analytics and machine learning come into play.
What excites you most about predictive analytics and machine learning?
Having lived my career in the software development field, I am excited to see a move away from millions of lines of hand-coded software, towards self-learning models that are significantly smaller and more intelligent.
What’s encouraging organisations to think more about predictive analytics and machine learning?
Organisations today are collecting larger and richer volumes of data more frequently. It’s close to impossible for them to analyse this data manually. This has been a driving force behind the use of technology such as predictive analytics and machine learning.
These technologies have matured in recent years and are now being packaged into more easily consumable offerings. It’s more achievable and cost effective than ever.
What insights can be uncovered with predictive analytics and machine learning?
Predictive analytics and machine learning can uncover data insights almost immediately that would otherwise take people days or even months to discover and understand.
If patterns in the data are regular and reoccur with a significant frequency, machine learning can ‘learn’ the patterns and make predictions based on this data.
For example, when vibration readings on the driveshaft of a machine increase, this is strongly correlated to bearing or shaft degradation. Monitoring vibration readings by collecting sensor data on the driveshaft and sending that data to an algorithm can identify degradation and predict potential failure.
What role does data quality play?
Data quality is crucial to any pattern recognition and prediction exercise. As a starting point, it’s important to ensure data is clearly understood, cleansed where necessary, correctly mapped and, in some cases, augmented or transformed. Machine learning models are then built and refined around this data.
As data volumes increase, there may be a greater propensity for models to tolerate some poorer quality data. As long as you have access to sufficient ‘good’ data points, the algorithms will still produce accurate and reliable predictions.
What else might organisations need for successful implementation?
It’s important to ensure a scalable and high-performance data platform is selected. It should be highly responsive and easy to integrate. Many cloud-based platforms provide solutions, each with its own benefits and drawbacks. Organisations need to understand these and align the solution with their key business drivers.
Our Trakka® 4.5 software contains a data platform at the heart of the system that can not only ingest, curate and analyse data from almost any source, but also unlock insights along with recommended actions for remediating issues. Trakka® 4.5 harnesses the power of machine learning and predictive analytics, and also manages and integrates perfectly into existing workflow processes. This results in more reliable equipment, fewer breakdowns, lower overall maintenance cost and longer asset life.
Which sectors are leading the way with predictive maintenance?
We have been helping customers for more than 25 years employ predictive maintenance techniques, particularly in the mining, oil and gas, rail and wind industries.
We don’t see any specific sector lagging to a great degree. But lower resourced companies that are marginally profitable and highly susceptible to their markets are further behind the big players who have access to more capital and resources.
In the mining world, Rio Tinto and BHP, for example, are progressing well on this journey, whereas the second or third tier miners have been a little slower to realise the potential.
What is the most common challenge facing DINGO’s customers?
Unexpected breakdown is the most common challenge in the early stages of our engagements. Our programs are geared towards bringing overall fleet condition back to an ideal health state within the first three months.
These early efforts reduce the number of unexpected breakdowns significantly and are heavily interconnected with our predictive maintenance processes and products.
What will machine learning and predictive analytics look like in ten years?
My prediction is that models will continue to improve and will allow a lot more human expertise to be embedded by subject matter experts, whereas now they are heavily influenced by data scientists.
I also predict platforms will evolve and commoditise almost every machine learning model to a point where users will be able to upload data, train and refine the models themselves without needing prior experience and data science expertise.
I believe machine learning and AI will become embedded into almost every backend system and app ecosystem out there. This will be driven by a need for software and apps to provide a more personalised user experience. This will come from analysing more of the user’s data and their specific circumstances.
In the meantime, DINGO will continue to drive progress in this rapidly evolving space by working closely with its customers and leading data science institutions, such as QUT, to develop highly advanced solutions with real world applications.
Chief Operations Officer, DINGO Software