PSI - Issue 54

Francisco Afonso et al. / Procedia Structural Integrity 54 (2024) 545–552 Francisco Afonso / Structural Integrity Procedia 00 (2019) 000 – 000

549

5

A defect-free template profile is first acquired, and all subsequent sensor acquisitions will be compared to it. This is accomplished through the combination of the custom Profile Analysis tool with a standard tool, Profile Template Matching. The latter enables saving a new template profile or choose a previously used one, outputting the difference profile which corresponds to the difference between the points in the current profile and in the template. The custom Profile Analysis tool was developed to receive the difference profile as input and analyze it, allowing the selection of the applied metric from the implemented ones, define their tolerances and the region of the profile that will be analyzed. During the analysis, the software allows the technician to consult the metric’s values, send a visual warning when a metric exceeds its tolerances and communicate such occurrences to a maintenance platform. Five metrics were implemented, the maximum difference (MD), average (AVG), median (MED), the difference between the average and the median (MM) and its absolute value (AMM). MD is calculated by simply identifying the highest absolute value in the profile, which corresponds to the value of the highest difference point between the current acquisition profile and the template, the value of this point is then identified to define if it is positive or negative. AVG is the result of dividing the sum of all difference profile points by the number of difference profile points, MED sorts the difference profile’s points from lowest to the highest, before identifying the middle value in the dataset , MM is calculated by subtracting the value of MED from AVG and AMM is the absolute value of MM. It is also possible to configure several output signals, each related to a metric. An analog signal was configured and processed by an external circuit which communicates with the maintenance platform via MQTT, indicating the severity and duration of its respective detected surface defect metric. 3.3. Trials A trial was conducted using this system and a printed model of a railway track containing a gradually increasing surface defect. The track model contains nine marks, from -1 to 7, where 0 is the last mark before the start of the surface defect which grows gradually from mark 1 to mark 6. The template profile was acquired in the -1 mark and the values of all the metrics were registered for each position, this process was repeated two more times, totaling three trials, using the same template profile obtained in trial 1. In Fig. 3, a relevant selection of the results can be found, where an error occurs when the sensor returns to the -1 mark while using a straightforward metric such as MD. This error is controlled by the use of a metric such as MM.

Fig. 3. Results for the railway track model test from mark -1 to 7, considering (a) MD; (b) MM.

A final test was also conducted where the system was attached to the undercarriage of a fully assembled train, in order to validate the support structure geometry and sensor solution, as well as the software acquisition and processing. 4. Wheel defect detection The second system, regarding wheel defect detection, was designed to be implemented in a maintenance context; thus, it may undergo its analysis on a partial structure of the train. As mentioned, a triangulation laser line sensor is