Background noise assessment of low-cost vibration sensors in precision manufacturing applications

Accurate, reliable, and consistent vibration measurements in manufacturing machines such as machine tools, form the basis for predictive maintenance and condition monitoring applications Recent advancements in Micro Electro-Mechanical Systems (MEMS) has led to a rapid adoption of low-cost industrial grade MEMS vibration sensors in contrast to traditional high-cost Integrated Electronics Piezo-Electric (IEPE) accelerometers which have been typically used in precision manufacturing setups. However, low-cost MEMS accelerometers are subject to various deterministic and stochastic noise phenomenon which often limits their performance especially when subject to low acceleration conditions. Preliminary work has been performed for characterization of baseline errors and uncertainties of such low-cost triaxial vibration sensors to evaluate their viability in metrological applications for Industry 4.0. However, employment of low-cost MEMS sensors for long term vibration measurements requires assessment and quantification of time dependent progression of the noise and its effect. Methods such as Allan Variance which are based on regression analysis of entire time-domain sensor data provide effective modelling and systematic assessment of background noise in MEMS based sensors in accordance with IEEE 1293-2018 standard.
To evaluate the background noise in these low-cost industrial MEMS sensors, continuous long-term data was recorded on a vibrationally stable test bed while establishing traceability according to the ISO 16063-11:1999 and ISO 16063-21:2003 standards. Linear accelerometer analysis was conducted to quantify and characterize various types of noise and random effects contributing to the sensor measurements while ensuring input to the setup is lower than intrinsic noise of the sensors. This work attempts to model noise parameters of low-cost MEMS vibration sensors to mitigate the baseline errors and random noise for employment in industrial manufacturing setups and smart condition monitoring applications. Results from this study will offer an improved framework for sensor stability through obtaining a low noise floor by a careful consideration of underlying random processes within metrology applications.