Predictive Maintenance In Social Housing:
Machine Learning

Predicting future maintenance events or asset failures by applying machine learning. Bimdl’s trained ML models allow organisations to predict the future by learning from the past events in existing data.


Predictive maintenance is one of the most highly desired outcomes from machine learning. Asset downtime is an issue that can cause huge disruption, predicting the potential failure and mitigating such events is “the Holy Grail of asset management”.

When we are entirely dependent on delicate or complex assets, system failure can affect the entire lifecycle and connected supply chain. This failure can also lead to significant direct costs and lost productivity for an end organisation whose success wholly depends on asset up time.

Imagine if we could accurately predict these failures and adapt corrective actions or mitigative measures.

In this post, “Predictive maintenance in social housing” we talk about Predictive Maintenance systems (PMS), how these systems can predict failures, corrective human intervention to avoid such failures, and the advantages of adapting business processes to include predicted outcomes.

Let’s get started 

What is Predictive Maintenance?

The main promise of predictive maintenance is to allow convenient scheduling of corrective maintenance, and to prevent unexpected equipment failures.

In essence, predictive maintenance relies on historic events and the known variables about a system, process or function relating to an asset, be it mechanical plant equipment or a residential fire door. Every physical component has a shelf life and when we start to analyse the spatial drivers local to an asset, we can start to recognise influencing factors that lead to down time.

We use predictive maintenance everyday, when our cars prompt us to book in for a service, or when we receive mobile prompts to charge our phone battery. These examples work well when you are local to the asset, or when the failures can be metered to provide triggers to intervene. But what about assets that don’t present obvious symptoms or that can’t easily be observed due to their wide distribution? 

Enter Machine Learning.

Machine Learning Output

How it works

The Machine Learning process begins with providing training data (historic repairs and maintenance data) into a selected algorithm, in this example a Predictive Maintenance algorithm. Normally, 20% of the available data is initially used to train the model and identify events in the data defined by the parameters of the algorithm. To test that the algorithm works correctly, new input data is fed in and the prediction and results are then checked.

If the prediction is not as expected, the algorithm will re-train multiple times until the desired output is found. This allows the algorithm to continually learn on its own and produce the most reliable outputs that will increase in accuracy over time.

Ingredients to begin?

The only ingredients needed to benefit from this approach is a Bimdl account and some historic repairs data.

The strength and reliability of machine learning outcomes is heavily dependant on the data used to train the algorithm. In essence we can’t predict the future unless we have documented the past. The more training data we feed machine learning algorithm, the more they can learn.

This is exactly why social landlords are a perfect fit to apply this module of Artificial Intelligence, most have been archiving data relating to repairs and maintenance for decades. Even where the format and structure of data is abnormal, well trained machine learning models can wrangle values to establish an alignment with the training models.

What are you waiting for……your next expensive reactive repair?

Start training your ML model today

You don't need all the data to begin. Simply provide a sample of your historic repairs data and bimdl ML will leverage our broader property data set to strengthen your model.  

Submit sample dataSign up here