Rail vibrations from draisine passages with speed information

# Rail Vibration Measurements - Dataset README Last updated: 2026-04-23 Corresponding Paper: An Experimental Dataset of Rail Vibrations for Sensor Mounting Analysis This dataset contains rail vibration measurements recorded in Westerstede using a draisine or small rail vehicle. The measurements were collected on 2025-10-13 and 2025-10-14 with four sensor positions. A regular vehicle pass is stored as one folder containing four WAV files, one file per sensor position, and one TXT metadata file. The dataset root contains four main data folders: ```text.|-- Day1ComparisonRun1/|-- Day1ComparisonRunWithAdditionalFixings/|-- Day2_magnetMountingOnly/`-- special_cases/``` ## Summary - Regular complete measurements with valid speed metadata: 93 passes- Complete measurements without valid reed-pair speed detection: 25 passes- Regular incomplete measurement folders: 39 folders- Incomplete measurement folders without valid speed detection: 22 folders- Special-case files: 49 WAV files and 12 TXT files- WAV format of the regular measurements: PCM, 16 bit, 3 channels, 1600 Hz In this README, "complete" means structurally complete: four WAV files and one TXT file in the same pass folder. This does not guarantee that the measurement is physically plausible or suitable for a specific analysis. In particular, speed values should be checked before using them for model training or quantitative evaluation. ## File Formats ### WAV Files Each WAV file contains the triaxial vibration signal of one sensor: - Format: RIFF/WAVE, Microsoft PCM- Resolution: 16 bit- Channels: 3, corresponding to the sensor x/y/z axes- Sampling rate: 1600 Hz- Duration of regular measurement folders: folder names end in `_10s`, indicating 10-second measurement windows On day 1, the sensors were mounted on a test carrier. The sensor axis orientation differs from day 2. Direct comparisons between day 1 and day 2 therefore require checking or transforming the axis orientation. ### TXT Metadata Each pass folder contains one TXT metadata file. The file name is associated with one sensor name, but the metadata describes the vehicle pass as a whole. Typical fields are: ```texttimestamp=2025-10-14 08:29:32 (Europe/Berlin)direction=B->Adt_ms=978.85distance_m=1.970speed_mps=2.013speed_kmh=7.25speed_cms=201.3``` For measurements without valid speed detection, the speed is set to `0.00` and the file contains a note such as: ```textnote=no valid reed pair detected for this pass``` ## Regular Measurement Folder Structure A typical complete day-1 folder: ```textDay1ComparisonRun1/output/complete_Measurements/20251013_132800_10s/|-- adhesive_20251013_132804_21.txt|-- adhesive_20251013_132804_21.wav|-- bolt_20251013_132804_15.wav|-- glue_20251013_132804_19.wav`-- magnet_20251013_132804_20.wav``` A typical complete day-2 folder: ```textDay2_magnetMountingOnly/output/complete_Measurements/20251014_082920_10s/|-- FlatMagnet_20251014_082925_3.txt|-- FlatMagnet_20251014_082925_3.wav|-- PointMagnet1_20251014_082925_0.wav|-- PointMagnet2_20251014_082926_0.wav`-- PointMagnet3_20251014_082927_1.wav``` File names contain the sensor or mounting label, timestamp, and a running ID. Pass folders were created by time-based clustering, so the four sensor files in one pass can have slightly different timestamps. ## Main Dataset Folders ### `Day1ComparisonRun1/` First measurement run from 2025-10-13. The sensors were mounted on a test carrier. Sensor positions are encoded in the file names as `adhesive`, `bolt`, `glue`, and `magnet`. The previous German draft described this run as using the normal screw tightening torque; this information is not directly encoded in the files. Structure and contents: - `output/complete_Measurements/`: 36 complete passes with speed metadata- `output/complete_Measurements/withoutSpeed/`: 7 complete passes without valid reed-pair speed detection- `output/incomplete_Measurements/`: 13 structurally incomplete measurement folders- `output/incomplete_Measurements/withoutSpeed/`: 2 incomplete measurement folders without valid speed detection Directions in the 36 regular complete measurements: - `A->B`: 17- `B->A`: 19 Speed range according to the TXT metadata: 4.75 to 385.79 km/h. Two values in this folder are clearly outside the expected range (`156.53` and `385.79` km/h) and should be checked or excluded before using speed as a target variable. ### `Day1ComparisonRunWithAdditionalFixings/` Second measurement run from 2025-10-13 with additional or modified fixation on the test carrier. Sensor and mounting labels are the same as in the other day-1 run (`adhesive`, `bolt`, `glue`, `magnet`). The previous German draft stated a defined tightening torque of 70 Nm for this run; this information is not directly encoded in the files. Structure and contents: - `clusters/`: empty in the current package- `output/complete_Measurements/`: 15 complete passes with speed metadata- `output/complete_Measurements/withoutSpeed/`: 6 complete passes without valid reed-pair speed detection- `output/incomplete_Measurements/`: 7 structurally incomplete measurement folders- `output/incomplete_Measurements/withoutSpeed/`: 4 incomplete measurement folders without valid speed detection Directions in the 15 regular complete measurements: - `A->B`: 11- `B->A`: 3- `forward - 2 person`: 1 Speed range according to the TXT metadata: 4.45 to 15.00 km/h. ### `Day2_magnetMountingOnly/` Measurement run from 2025-10-14 with direct magnetic mounting on the rail. One sensor is labeled `FlatMagnet`; the other three are labeled `PointMagnet1`, `PointMagnet2`, and `PointMagnet3`. Structure and contents: - `clusters/`: empty in the current package- `output/complete_Measurements/`: 42 complete passes with speed metadata- `output/withoutSpeed/`: 12 complete passes without valid reed-pair speed detection- `output/incomplete_Measurements/`: 19 structurally incomplete measurement folders plus two loose WAV files- `output/incomplete_Measurements/withoutSpeed/`: 16 incomplete measurement folders without valid speed detection Directions in the 42 regular complete measurements: - `A->B`: 16- `B->A`: 26 Speed range according to the TXT metadata: 5.45 to 12.06 km/h. Besides the actual measurement folders, `output/complete_Measurements/` also contains analysis and machine-learning artifacts, for example `runs_all/`, `runs_extended/`, `prepared_datasets/`, `cnn_mfcc_out/`, `firstTryCatBoostRegressor/`, and additional result folders. No Python scripts, notebooks, or `.npz` files were found in the current cleaned package. Earlier references to an included `build_dataset_min.py` therefore do not apply to this package version. ### `special_cases/` Special measurements from 2025-10-14. These data are not part of the regular standard dataset. They are intended for studying specific vibration and vehicle-operation scenarios. These folders do not necessarily contain exactly four WAV files plus one TXT file. Subfolders: - `hammer_blows_10m_before_magnet_1/`: hammer blows 10 m before magnet 1- `hammer_blows_5m_before_magnet_1/`: hammer blows 5 m before magnet 1- `hammer_blows_5m_before_magnet_1_opposite_track/`: hammer blows 5 m before magnet 1 on the opposite track- `hammer_blows_5m_after_magnet_3/`: hammer blows 5 m after magnet 3; contains an additional `pollution/` folder with noisy or contaminated extra files- `approach_brake_and_slow_pass/`: approaching, braking, and slowly passing the sensors- `driving_on_left_side/`: driving on the left side; the current package contains only one WAV file in this folder- `pushed_over_sensors_without_running_over_them/`: draisine pushed across or past the sensors without a regular pass- `calibration_and_between_runs_mostly_invalid/`: intermediate, calibration, and mostly invalid measurements ## Recommended Use For standard analysis and machine-learning experiments, start with the regular complete measurement folders: ```textDay1ComparisonRun1/output/complete_Measurements/20*_10s/Day1ComparisonRunWithAdditionalFixings/output/complete_Measurements/20*_10s/Day2_magnetMountingOnly/output/complete_Measurements/20*_10s/``` For supervised learning with speed as the target variable, exclude `withoutSpeed` folders unless a different target variable is defined. Incomplete measurement folders and `special_cases/` can be useful for additional analyses, but they should be checked manually first. For comparisons between day 1 and day 2, at least the following points must be considered: - Different sensor and mounting labels between day 1 and day 2- Different day-1 sensor axis orientation caused by the test carrier- Different mounting conditions: test carrier on day 1, direct magnetic mounting on day 2- Possible outliers in the automatically calculated speeds