Structural Monitoring Without a Budget-Laboratory Results and Field Report on the Use of Low-Cost Acceleration Sensors

Abstract

Authorities responsible for critical infrastructure, particularly bridges, face significant challenges. Many bridges, constructed in the 1960s and 1970s, are now approaching or have surpassed their intended service life. A report from the German Federal Ministry for Digital and Transport (BMVI) indicates that about 12% of the 40,000 federal trunk road bridges in Germany are in "inadequate or unsatisfactory" condition. Similar issues are observed in other countries worldwide. Economic constraints prevent ad hoc replacements, necessitating continued operation with frequent and costly inspections. This situation creates an urgent need for cost-effective, permanent monitoring solutions. This study explores the potential use of low-cost acceleration sensors for monitoring infrastructure structures. Inclination is calculated from the acceleration data of the sensor, using gravitational acceleration as a reference point. Long-term changes in inclination may indicate a change in the geometry of the structure, thereby triggering alarm thresholds. It is particularly important to consider specific challenges associated with low measurement accuracy and the susceptibility of sensors to environmental influences in a low-cost setting. The results of laboratory tests allow for an estimation of measurement accuracy and an analysis of the various error characteristics of the sensors. The article outlines the methodology for developing low-cost inclination sensor systems, the laboratory tests conducted, and the evaluation of different measures to enhance sensor accuracy.

Keywords: MEMS accelerometers; SHM; inclination sensors; low-cost sensors; structural monitoring.

Figure 11

Left and centre: inclination changes due to temperature variation while sensors are being static at rest // Right: sensor temperatures and climate chamber temperature.Interestingly, despite sensor 4 maintaining almost ideal constant temperature due to temperature control (see right plot in Figure 11 in purple), its x-axis inclination values still exhibited changes (see left plot in Figure 11 and Figure 12). With the sensor assumed to be at rest and at a constant temperature, such variation is not expected. Conversely, y-axis inclination remained nearly stable, apart from slight temporal drift during the trial; see centre plot in Figure 11. Given the correlation between x-axis pseudo-inclination change and temperature (correlation factor −0.93), it is likely that thermal differences within the board (up to 90 °C between sensor and surroundings) induced mechanical stresses, causing real inclination via thermal strains—possibly a single-axis board warp from these stresses.

Figure 12

Relation of calculated inclination of sensors at rest with temperature variations in climate chamber at temperature-controlled phase of experiment.

Figure 1

A 3 axes acceleration sensor in relation to g, y-z plane, in case $a_x=0$ becomes $a_z^{\mathrm{\prime }}=a_z$, in case $a_x\ne 0$ becomes $a_z^{\mathrm{\prime }}=\sqrt{a_x^2+a_z^2}.$

Figure 2

Schematic representation of displacement measurements with an accelerometer.

Figure 3

PCB with 10 MEMS accelerometer (MPU-9250) and Nivel210 as reference.

Figure 4

RMSE of mean sensor data vs. number of sensors being used.

Figure 5

In total, 50 MEMS accelerometers on five PCBs stacked over each other (left) and one PCB with 4 MEMS accelerometers (right).

Figure 6

Schematic representation of data processing and storage chain.

Figure 7

Frequency distribution of bias and scale factors of 50 MEMS accelerometers.

Figure 8

Drift of sensor data over time.

Figure 9

Schematic representation of system configuration.

Figure 10

Experimental setup in a climate chamber: four MPU-9250 sensors on the top-layer PCB, connected with a flat cable to the processing unit at the bottom layer.

Figure 13

Acceleration vs. sensor temperature for x-axis (moving average values); left: 14 days data; right: only first 7 days with temperature control.

Figure 14

The x-axis acceleration measuring data of temperature-controlled sensors; (left): with least squares linear fit as an approximation for the temperature error; (right): temperature compensated.

Figure 15

Examples of improved acceleration and inclination data by forming the arithmetic mean of all sensors.

Figure 16

(left): relation between temperature gradient and sensor output (or resulting inclination, respectively); (right): heating phase (red) and cooling phase (blue) of the temperature-controlled part of sensor 2.