Process Quality Monitoring Through a LSTM Network Derived from a Rule-Based Approach

The railway infrastructure condition is a crucial factor for thesafe and efficient operation of trains. Regular maintenanceis inevitable as the track geometry degrades over time dueto traffic and environmental effects. To restore the ideal positionand provide sufficient durability of ballasted track socalled tamping machines are used. These machines lift thetrack, correct the longitudinal level and the alignment of thetrack panel and tamp the ballast. During the tamping processthe tamping tines penetrate the ballast bed, fill voids and compactthe ballast underneath the sleepers by a squeezing movementwith superimposed vibration. A detailed description ofthe tamping cycle can be found on section 2. Monitoring andevaluating this tamping process is essential for maintainingprocess quality. This can be achieved through a variety ofsensors, such as incremental encoders, angle encoders, temperature,pressure, and acceleration sensors, coupled with ameasurement unit (DAQ and edge device) to collect, locallystore and transmit the data to a cloud. This paper explores thedevelopment of a rule-based algorithm for assessing the qualityof the tamping process execution in reference to its nominalchronological sequence. The focus is on identifying tampingoccurrences and classifying them into acceptable (OK)or non-acceptable (NOK) categories. This involves selectingrelevant measurement parameters and processing them,considering the inherent imprecision in real-world processes.Empirical thresholds are established to differentiate betweengood and bad outcomes. The classification approach has tobe sufficiently generic in order to cover a high variety of customizedtamping machine types. As each machine is individuallydesigned, the process of generalization is challengingand complex. The paper demonstrates the accuracy and universalapplicability of the developed rule set across differenttamping machines. The model’s effectiveness is validated usingthe Hold-Out-Test-Set method. Furthermore, the rule-set-achieved outcomes are compared with results gained from anLSTM network. Both the rule-based approach and the neuralnetwork demonstrate precision, but the latter requires significantlymore effort.