
Thanks to the results of the technology using artificial intelligence to process sound and vibration signals, in cooperation with the Prague Public Transport Company and Wiener Linien, we were able to identify opportunities for extending the life of escalators, reducing their downtime, and prioritizing the activities of the maintenance team.
Over a year-long collaboration on escalator monitoring through Neuron Soundware solutions has culminated in controlled stress testing of individual escalator components located in a subway transfer station. The test always aimed to verify the detection of an impending fault. This case study will guide you through the testing and show you the results.
The monitoring of escalators as specific equipment has resulted in the adaptation of technology and procedures that NSW normally uses in the field of industrial enterprises (monitoring pumps, fans, compressors, etc.). This case study also served to build segment expertise in the field of escalators in their entirety, not just individual sub-components.
Thirteen machine parts on the left part of the escalator were selected to simulate failure conditions. These were the monitored type of escalators Thysen FT-732.
Table: list and types of simulations together with the escalator part to be fitted
The change in sound and anomaly detection using a mathematical model of the spreading material on the ridge plate, the loss of oil buildup in the worm gearbox and gear train, the change in the adjustment of the relief curves, and the loosening of the handle proved conclusive. If at least a weak detection occurred, the model upgrade would presumably improve the result significantly (rollers, tracks).
This analysis used a neural network (link), which was trained on data from all 21 machines previously fitted with NSW sensors in collaboration with DPP.
Before the audio signal is processed by the neural network, the audio is converted into spectrograms via STFT (Short Term Fourier Transform). A spectrogram is a representation of the intensity of an audio signal at different frequencies over time.
The y-axis represents the frequency of the signal, the x-axis represents the time, and the coloration rate represents the intensity of the signal – the loudness. (Yellow = highest energy, signal intensity, Dark purple = low energy, low signal intensity).
To illustrate, the top image is the pure sound signal, the middle is a spectrogram representation of that signal, and the bottom shows the frequency distribution for the nominal sound (green curve) and the sound simulation of the damaged machine part (brown curve).
The crest plate is the element of the escalator that is located at the beginning and end of the escalator. It serves as a transition space between the staircase and the escalator board.
While the escalator is in operation, the spreader material becomes jammed under the crest plate, causing the stair treads to grind out, and in the worst-case scenario, the teeth may break out of the crest plate. This type of failure can escalate very quickly from the beginning of the problem to its fatality, making it extremely important to implement real-time monitoring and to perform the measurements frequently.
Spreading material trapped under the ridge plate that reaches further into the escalator can cause costly consequences such as incorrect adjustment of the stair guide ruler and its subsequent total misalignment.
The sound was distinctive even to the human ear. There was a clear difference in the recordings compared to normal operation as evidenced by the spectrogram and frequency spectrum. These manifestations were then reflected by the model that detected the fault.
As can be seen from this fault analysis, the model clearly detected the fault. When compared to normal operation without the fault, with people walking on the escalator, the model was able to distinguish the difference and did not generate false alerts.
As noted in the first table, there are other failures apart from crest plate alien object detection. The neural networks successfully detected a lubricant leakage and untensioned handlebar as visualized in the images bellow:
The handle pressure is used to properly tension the handle belt so that it does not slip or fall due to excessive slack and is not overtightened or cause increased wear.
This case study included four simulations that demonstrated the suitability of remote acoustic monitoring solutions using artificial intelligence data processing technology. The benefits that emerged from the testing are as follows:
Maintenance of transport equipment can instantly display information about the operation of a specific monitored part of the escalator.
The project introduces Transport Facilities Maintenance to the monitoring system and the development of internal processes for working with the escalator monitoring data.
The project contributes to learning about the requirements for installing sensor technology in a metro operating environment.
The project is helping to identify escalator components suitable for the installation of acoustic data acquisition technologies. The information obtained is used to assess and select machine components suitable for monitoring by acoustic detection in the future.
The project helps to outline a way to manage the constraints arising from the shortage of skilled maintenance workers.
DPP cooperates with Neuron SW in the field of continuous monitoring and early detection of defects on 21 escalators as well as extensive simulation and early detection of real defects. The collaboration is very close across the levels of both entities and given the commitment of both teams, beneficial progress is being made in quick succession.
Our team worked with Neuron Soundware to identify key locations for sensor placement on traffic devices in order to extend the interval of preventive checks based on the results of the testing. Other benefits include reduced operational downtime, extended escalator lifetime, and prioritization of maintenance team activities. Specific quantifications of the overall benefits are the subject of an ongoing study.
Comprehensive monitoring covers the operation of the escalator motor and gearbox, the condition of the crest and platform plates, the tension status of the pull chain and the position of the tensioning trolley, and the tension status of the moving handrails, and the positioning of the relief curves. For these escalator components, the Neuron SW system reliably detected deviations from the nominal state and alerted the emerging problems in their initial phase.
The technical solution from Neuron Soundware has proven its functionality and therefore we plan to continue our activities with Neuron SW in the field of digital monitoring of escalators.
Ing. Petr Vondráček, Head of Transportation System Service, Traffic Route Metro, Dopravní podnik hl. m. Prahy, a.s.