How To use AI in Downtime Prediction - AiRobot

When it comes up to inte­grating AI within already working core processes and busi­ness, one can be struck by fear and be ret­i­cent to change. A simple and harm­less step towards reaching full automa­tion is to monitor pro­duc­tion pipelines and store data. The next step involves con­structing dash­boards con­taining accu­rate metrics and main­te­nance pre­dic­tions, thus achieving AI in Down­time Prediction.

AI in Data Forecasting

By tracking cir­cum­stances, data sta­bility and applic­a­bility, the insights become more rel­e­vant and remain sig­nif­i­cant for much longer. Being able to accu­rately predict future data allows for deeper delving into data analysis. The end result is the pro­duc­tion of much more qual­i­ta­tive busi­ness insights and dash­boards. Knowing some of the future data might help in ame­lio­rating or even fully pre­venting big cat­a­stro­phes waiting around the corner to happen.

Having access to a live data stream of certain metrics (such as voltage) creates a suit­able envi­ron­ment for building AI that fore­cast future values of that metric. 

AI in Data Forecasting
Voltage Fore­casting (A)

In figure A, two graphs are put head to head in order to demon­strate the usability of an AI used for Voltage Fore­casting. The left side of the plot con­tains past values of voltage. A voltage fore­casting AI pre­dicts future pos­sible values with a margin error, dis­played on the right part.

AI in Downtime Prediction

A machine is not perfect, and even­tu­ally it will stop func­tioning and require main­te­nance. Machine down­time periods can be very hurtful to busi­nesses. The costs can go to very large amounts, espe­cially when the main­te­nance process will take a few months.

Building a large and broad dataset of infor­ma­tion about a spe­cific machine usually means aggre­gating some spe­cific machine data, metric time data and fore­casts. This infor­ma­tion can be further aug­mented with past main­te­nances periods in order to train an AI to predict future down­time risks.

AI in Downtime Prediction
Down­time Risk Pre­dic­tion (B)

The red peaks from the left part of the figure (B) rep­re­sent past crit­ical errors which put the machine on down­time. The orange values, from the right, rep­re­sent future risk values of crit­ical errors hap­pening. A down­time pre­dic­tion AI pro­duces the risk values from pro­cessing already known data.

The graphs from above are con­stantly updating and moving with the time­line in a live sce­nario. Fully automating the down­time pre­dic­tion system will require an error threshold in order to trigger a message or an alarm.


There are many ways in which com­pa­nies adapt to the new stan­dard of Industry 4.0. Many of them can be very demanding and can be harmful to past systems. Harm­less alter­na­tives, such as pre­dic­tive main­te­nance, can be safely inte­grated within the system and rep­re­sent a strong first step.

Imple­menting an AI in Down­time Pre­dic­tion requires read access to some live data streams. This data has various appli­ca­tions in data science and ana­lytics. Some impor­tant ones are the processes building the com­po­nents nec­es­sary to fully auto­mate pre­dicting downtime.

By having an auto­mated pipeline for pre­dic­tive main­te­nance, the busi­ness will have lower pro­duc­tive losses and will require less human labor in order to function. 

Final words

In this blog post, the AiRobot team tries to reach out with its work. 

In the mean­time, more updates are in the works with an attempt at cov­ering as many AI use cases as possible. 

Make sure to follow the news!

Further readings

More infor­ma­tion about pre­dic­tive main­te­nance is avail­able here.

Recent posts from us: