Predictive Maintenance in the Production Chain

By utilizing predictive algorithms, we can identify a machine’s usual behavior and generate alerts the moment anomalies are detected. This enables targeted maintenance based on machine monitoring, providing greater insight and visibility into asset status.

Objectives

• Identify anomalous behaviors
• Perform targeted maintenance through active monitoring
• Increase production by reducing unforeseen downtime
• Understand machine status and study their behavior

Asset monitoring in the industrial sector is an increasingly prevalent reality. The criticality of their proper functioning, due to the cost of machinery and the losses caused by downtime, is becoming more important. It is crucial to be able to identify anomalous behaviors to prevent breakdowns, as well as to support decisions on when and where maintenance is needed.

What we do

We identify and monitor which assets should be analyzed, and what sensorization and variables are being collected. Through an exploratory analysis, we identify those that can provide value and the granularity requirements. We select the most suitable algorithm and define the parameters that will generate alerts, as well as the implementation of all automation and reporting.

How We Do It

Exploratory Analysis

We analyze data quality and correlations between variables with the aim of identifying the most relevant ones, corroborating this with the client’s business knowledge. This is done using correlation matrices, mutual information, various graphs, or intermediate dashboards.

Algorithm Selection

We employ decision trees (XGBoost, LightGBM) or neural networks (RNN, LSTM). After generating models capable of predicting variable information, we characterize these models to identify normality. This allows us to verify if the behavior matches the pattern or is anomalous.

Data Exploitation

We create a dashboard to visualize results, configured and adapted to client requirements. We implement the desired reporting system for alerts (emails, database entries, or their PI system), adjusting it to specific needs.