PSI - Issue 38

Frédéric Kihm et al. / Procedia Structural Integrity 38 (2022) 12–29 Kihm, Miu, Bonato / Structural Integrity Procedia 00 (2021) 000 – 000

14

3

Engineers play a fundamental role in the DAQ phase because their expertise is needed to maximise the quality and quantity of data acquired in the most efficient way. Be either a road field test or a wind tunnel test, vehicle measurements are expensive. To perform such tests, one must first build a prototype or buy or rent the selected car (for component suppliers). The required sensors need to be mounted, connected, and wired. There must be enough room for all sensors, cables, and a comfortable environment for both the driver and the acquisition system. Additional care must be taken when measuring the response of a component, which needs to be equipped with sensors (e.g. strain gages and thermocouples) by a reliable supplier, and then assembled in the vehicle. Because of tight time and budgetary constraints, the key role of the test engineer is to make sure that the measurement sessions are not jeopardized because of improper sensor mounting or inadequate data acquisition and storage. Sensors need to be tailored to the physical value of interest. It is therefore fundamental to have some a priori knowledge of the value range (maximum and minimum value, the frequency, etc.). Sensors must also be accurate, enough sensitive to properly measure small variations, but robust toward the inevitable environmental stress resulting from driving ground: shock, heat, humidity, and contamination associated to the potential adverse conditions of the road profile (dust, mud, water, etc.). Similarly, the acquisition system must be tailored to the measurement type. There must be a trade-off between system performance and robustness and in some case its dimensions. As a typical example, integrated circuit piezoelectric (ICP) accelerometers are less intrusive (smaller and lighter) and more accurate than capacitive ones, but less resistant to high temperature and shock. ICP sensors would perfectly fit for vibration measurements on the chassis or cabin component, but be unreliable for engine vibrations, due to the high temperature reached during combustion.Moreover, the electromagnetic compatibility of the acquisition systems and logger should be verified to avoid the presence of electromagnetic parasitic noise. Finally, the test engineers consider the next phase of the task, which is the data post processing and analysis. It is therefore fundamental to work with properly labeling of measurements channels and data saved in an exploitable format (Bracke et al (2018)). 2.2. Data cleaning The collected data is usually represented as time series, either with evenly spaced time steps – with a given sample rate, or with unevenly spaced time steps, also known as time stamped data. Data cleaning and filtering is a mandatory step in this process, as erroneous and bad data will significantly impact the quality and robustness of any downstream analysis. Digital bus data raises a number of challenges in terms of quality, consistency, lack of certain data, etc. For instance, data points that are undefined or unrepresentable using numeric values, referred to as NaN or “Not a Number” need to be addressed before doing any further calculation. Time stamps when such NaNs occured can be ignored entirely or the NaNs can be replaced by a numerical value -this could be a fixed value or the last available numerical value in the stream or based on interpolation, etc. With strain gage data, the most current issues include spikes, drifts, offsets and noise. The engineer’s judgement is often required to decide what is effective data and what is noise or whether a given spike is a physical peak or the result of some electromagnetic interference? Off-the-shelf engineering tools exist to assess and clean time series of measured data. This can be based on digital filtering, running median filters, interpolations, etc.

2.3. Data reduction

The next phase involves reducing the time series data into metrics. Those metrics can be as simple as averages or maximum values. Metrics can be calculated over fixed periods of time and cumulated, counted and classified into histograms. More advanced metrics can be calculated based on a combination of sensor data and imported metadata (vehicle type, driver profile, etc.). Data scientists refer those to metrics as features. A feature is a measurable property that can be used for the analysis.

UNRESTRICTED