HHLA is using artificial intelligence (AI) with different focal points and characteristics in a number of projects in order to test new possible applications. A particularly interesting application in terms of cost efficiency is predictive maintenance, which is the creation of reliable predictions on service life and expected damage to equipment.
Let’s take steel cables as an example – these are subjected to heavy daily use on HHLA’s container gantry cranes. In 2019, 138 of these cables with a length totalling 113.2 kilometres had to be replaced at HHLA Container Terminal Burchardkai alone. But when exactly do they need to be replaced, or at least inspected? Finding the right timing for this can reduce the associated costs of maintenance and replacements.
Until now, the steel cables have been manually inspected at regular intervals and the data collected compared with normalised threshold values. The comparison then shows whether the cable needs to be replaced. The timing of the replacement can, however, be at an inopportune moment and coincide with the unloading of a ship. This results in additional costs and delays in operations.
The service life of the cable can also vary greatly, and the cables are purchased in advance in the expected quantities. They must be stored until the replacement date, occupy valuable terminal space and can be damaged even before they are installed.
Therefore, HHLA Technik launched a project to be able to better predict the expected service life and thus the optimum replacement date. The team chose a suitable machine learning (ML) module and were given access to operating data on container gantry crane cable maintenance from the last six years to train a neural network. The objective: determine the replacement date two weeks in advance.
ML helps recognise patterns or regularities in existing structures, and neural networks are used to handle complex relationships between many variables in the available data. Identifying patterns in such complex data sets is one of the strengths of neural networks.
The analyses by HPC's ML experts indeed prove that the accuracy of the predictions matches the actual service life of the ropes very well after only a few fine adjustments of the neural networks. In the next step, the developed AI model will now be implemented by HPC in software that can be easily used by HHLA Technik and integrated into the maintenance planning of the terminals.
Ulf Bockelmann, Managing Director of HHLA Technik, thinks the fast results are “really impressive” and adds, “We need to look even closer at how ML can best be used in our industry. I could imagine, for example, aligning maintenance intervals depending on the load parameters. Based on the predicted service life, we will be able to plan maintenance work in a more operationally compatible way in the future. Ideally, we would further narrow down the causes of increased wear and derive countermeasures. But we’re still a good few steps away from that.”
Benjamin Heusser, a BSI machine learning engineer who consulted on the project, was very pleased, saying, “The combination of the HHLA project team’s industry insights and our ML expertise provided optimal conditions for a successful project.”