Industrial businesses of all shapes and sizes gather data in order to make sense of patterns and relationships so as to guide us towards better decisions and results in the future.

From analyzing business-wide environmental KPIs to the RPM of an engine in a single piece of machinery, we can learn something about performance and assess the different variables that are going into it in order to establish better ways of operating.

Sometimes these relationships can be seen plainly with visualizations on dashboards that humans can easily interpret, e.g. a machine is getting hot because we’re in the middle of a heatwave so it needs to be turned off, and there’s no need to investigate with more complex tools.

Often, however, further analysis is needed, e.g. this machine is getting hot but we’re operating within normal parameters – what’s going wrong?

When can you use a machine learning algorithm?

When you know how to approach a problem and are able to judge success vs. failure (i.e. there is a metric), you can create an algorithm that can find ways to improve the current status-quo. Algorithms are an unambiguous, finite sequence of well-defined computer instructions to solve a class of problems and are guaranteed to produce a result.

To generate the positive outcome, however, an algorithm needs to be implemented into a workflow.

Workflows are based on the traditional programming paradigm, which gives computers explicit instructions to achieve an outcome i.e. if this, then that. An example of this kind of analysis is the combination of energy consumption data with HVAC unit activity and performance to create insights into efficiency e.g. when the daily low is 19 degrees Celsius, it’s more efficient to run a HVAC overnight than to turn it off at 11pm and then back on again at 7am.

Using traditional workflows, it is possible to carry out even moderately complex analyses (such as descriptive statistics, statistical process control, etc.).

There are times, however, where we simply don’t know which variable(s) are affecting performance or what the effect of altering them would be on metrics.

It is in these circumstances that we would looking to use a use a Machine Learning algorithm. Machine Learning gives computers the ability to learn (extract patterns and insights) from data without being explicitly programmed to look for them, i.e. it can identify in big data sets the variables that are impeding performance.

To function, they need good data, information about actual outcomes, and some amount of human supervision to train it so that it knows what a good result is and can then relate it to practical steps that can improve it.

In short, you should use machine learning when you know what the business problem is, have or can get access to the data that goes into achieving it, and are able to judge success – but have no idea what or how to achieve it.

Within an industrial context, the best place for all this to happen and for you to apply machine learning is within an integrated data, AI + IoT platform. An integrated data, AI + IoT platform enables you to use IIoT devices, integrate data streams, create custom workflows, apply machine learning to them and generate visualizations that can guide people or create automations to deliver improved performance.

If you’d like help on knowing how to build a machine learning algorithm, then you can read more in this blog, or alternatively, if you’d like to explore how and where to get started with machine learning within your business, then speak to us today.