Preventive maintenance
What we do

Failures are expensive, so we’ve developed preventive maintenance algorithms.

Failures are expensive, so we’ve developed preventive maintenance algorithms that help you look ahead by detecting potential vulnerabilities in large-scale systems and processes, and they can track anomalies that might point to future problems. 

Algorithms of this type can estimate the so-called “remaining useful life” of products, assessing how long a machine is likely to run before it needs repair. Such intelligent estimates can bring substantial savings to any business.

Digica has extensive machine learning experience, from standard statistical algorithms to sophisticated deep learning solutions, according to what the project needs. They enable the system to flag up anomalies and assess the remaining life of the product, as well as find reasons for the failure and point to possible solutions.

Case studies
Preventive maintenance of telecom end-user devices

Preventive maintenance of telecom end-user devices


What the customer wanted to achieve

The customer has a large scale telecommunications infrastructure and various end-user devices like laptops, mobile phones, tablets, STBs, and intelligent home appliances connected to end routers. Certain network conditions and end-users’ devices were causing communication breakdowns. The customer wanted to be able to monitor this complex network to avoid problems for its customers.

How Digica helped the customer

We trained a neural network model to predict when a specific mobile device will crash in the near future. We trained it using a continuous stream of information, unique to every mobile device, consisting of:

  • Its internal state (OS, make, model, set of installed applications), and its user's behaviour (number and time of phone calls, number and time of text messages, WiFi on/off, etc.)
  • Additionally, the model was able to advise customer services on preventative maintenance actions such as upgrade/downgrade of OS, removal of applications etc.

What we have achieved

The model predicts mobile device failure with 91% accuracy.

Technologies used

SHAP, decision trees (XGBoost)