Acoustic sensors for predictive maintenance at sea

Challenge

Predictive maintenance has tremendous potential for reducing maintenance and downtime cost at sea. Maersk asked Flowtale to help reduce engine wear by identifying suboptimal lubrication events in main engine cylinders and optimising lubrication consumption.

Solution

Using acoustic sensor data from the main engine cylinders, Flowtale could detect wear, predict scuffing and determine whether lubrication levels were correct — with a convolutional neural network classifying suboptimal lubrication cycles.

• End-to-end structure. High-frequency sensors mounted on cylinders, data-gathering infrastructure established, and a CNN-RNN model built on spectral audio features.
• Proven in harsh conditions. The sensors function reliably in the machine-heavy environment of a working vessel and deliver clear, consistent data through a robust pipeline.
• Successful AI modelling. The gathered data was transformed into AI-suitable features, and the network predicts lubrication events with very high accuracy.

Results

• Sensors proven in the engine-room environment
• Reliable pipeline for large data volumes
• Impressive accuracy predicting low vs. normal/high lubrication