Low-Power Detection and Classification for In-Sensor Predictive Maintenance Based on Vibration Monitoring (original) (raw)
Related papers
Vibration Anomaly Detection using Deep Neural Network and Convolutional Neural Network
International Journal of Modeling and Optimization, 2021
Identifying the “health state” of the equipment is the domain of condition monitoring. The paper proposes a study of two models: DNN (Deep Neural Network) and CNN (Convolutional Neural Network) over an existent dataset provided by Case Western Reserve University for analyzing vibrations in fault diagnosis. After the model is trained on the windowed dataset using an optimal learning rate, minimizing the cost function, and is tested by computing the loss, accuracy and precision across the results, the weights are saved, and the models can be tested on other real data. The trained model recognizes raw time series data collected by micro electro-mechanical accelerometer sensors and detects anomalies based on former times series entries.
Low Power Anomaly Detection and Notification Systems using Deep Learning
2017
Considering the advancements in computer vision and deep learning algorithms, with facilitations of open source libraries aiding various applications, a lot of research work is done keeping in mind various problems faced by the society. Such two important aspects are security surveillance and environment wellness. With the use of state of the art machine learning algorithms, such problems can be solved. TensorFlow is one such eminent open source library used for training, classification and identification purposes, deployed to train a model that is able to classify assault objects such as knife, rifle, handgun etc. for security surveillance and raise an immediate alarm on confident identification. Similarly, the same model has been used and deployed on a portable computation chip, in this case, Raspberry Pi, to be installed on a UAV that will take aerial shots of city streets and send geo-location of the area with anomalies such as potholes and unattended garbage heaps. The model tr...
IEEE Transactions on Circuits and Systems I: Regular Papers, 2019
To overcome the energy and bandwidth limitations of traditional IoT systems, "edge computing" or information extraction at the sensor node has become popular. However, now it is important to create very low energy information extraction or pattern recognition systems. In this paper, we present an approximate computing method to reduce the computation energy of a specific type of IoT system used for anomaly detection (e.g. in predictive maintenance, epileptic seizure detection, etc). Termed as Anomaly Detection Based Power Savings (ADEPOS), our proposed method uses low precision computing and low complexity neural networks at the beginning when it is easy to distinguish healthy data. However, on the detection of anomalies, the complexity of the network and computing precision are adaptively increased for accurate predictions. We show that ensemble approaches are well suited for adaptively changing network size. To validate our proposed scheme, a chip has been fabricated in UMC 65nm process that includes an MSP430 microprocessor along with an on-chip switching mode DC-DC converter for dynamic voltage and frequency scaling. Using NASA bearing dataset for machine health monitoring, we show that using ADEPOS we can achieve 8.95X saving of energy along the lifetime without losing any detection accuracy. The energy savings are obtained by reducing the execution time of the neural network on the microprocessor.
Accelerometer-Based Event Detector for Low-Power Applications
Sensors, 2013
In this paper, an adaptive, autocovariance-based event detection algorithm is proposed, which can be used with micro-electro-mechanical systems (MEMS) accelerometer sensors to build inexpensive and power efficient event detectors. The algorithm works well with low signal-to-noise ratio input signals, and its computational complexity is very low, allowing its utilization on inexpensive low-end embedded sensor devices. The proposed algorithm decreases its energy consumption by lowering its duty cycle, as much as the event to be detected allows it. The performance of the algorithm is tested and compared to the conventional filter-based approach. The comparison was performed in an application where illegal entering of vehicles into restricted areas was detected.
Future Internet, 2021
In this work, we present an innovative approach for damage detection of infrastructures on-edge devices, exploiting a brain-inspired algorithm. The proposed solution exploits recurrent spiking neural networks (LSNNs), which are emerging for their theoretical energy efficiency and compactness, to recognise damage conditions by processing data from low-cost accelerometers (MEMS) directly on the sensor node. We focus on designing an efficient coding of MEMS data to optimise SNN execution on a low-power microcontroller. We characterised and profiled LSNN performance and energy consumption on a hardware prototype sensor node equipped with an STM32 embedded microcontroller and a digital MEMS accelerometer. We used a hardware-in-the-loop environment with virtual sensors generating data on an SPI interface connected to the physical microcontroller to evaluate the system with a data stream from a real viaduct. We exploited this environment also to study the impact of different on-sensor enco...
IEEE Open Journal of the Industrial Electronics Society
In recent years, many industrial facilities have been required to install sensing system to monitor the health status of the equipment. Although the PC-based diagnosis system is a good alternative, it undoubtedly increases the hardware cost greatly, and also causes restrictions on system assembly. In some critical environments, installation of the PC-based system is not suitable because of high temperature, dust, and humidity. On the contrary, a high mobility portable smart sensor will be relatively adequate for onsite inspections. To solve this problem, we developed a low-cost embedded handheld smart sensor. Instead of using sophisticated algorithms, certain frequently used statistic indicators are considered. Nevertheless, due to the resource limitations of the embedded systems, it causes difficulty for real-time realization and therefore, the recursive architecture of the statistic indicators is derived. These statistical factors are then fed into a Gaussian classifier for online defect detection, which gives a great contribution to field operators for onsite inspections. The highly integrated hardware/software co-design of the developed device provides userfriendly and high mobility for field inspections. Finally, a practical industrial application regarding the online diagnosis of a solenoid valve actuator defect is presented to demonstrate the effectiveness of the developed embedded smart sensor.
ADIC: Anomaly Detection Integrated Circuit in 65-nm CMOS Utilizing Approximate Computing
IEEE Transactions on Very Large Scale Integration (VLSI) Systems, 2020
In this paper, we present a low-power anomaly detection integrated circuit (ADIC) based on a one-class classifier (OCC) neural network. The ADIC achieves low-power operation through a combination of (a) careful choice of algorithm for online learning and (b) approximate computing techniques to lower average energy. In particular, online pseudoinverse update method (OPIUM) is used to train a randomized neural network for quick and resource efficient learning. An additional 42% energy saving can be achieved when a lighter version of OPIUM method is used for training with the same number of data samples lead to no significant compromise on the quality of inference. Instead of a single classifier with large number of neurons, an ensemble of K base learner approach is chosen to reduce learning memory by a factor of K. This also enables approximate computing by dynamically varying the neural network size based on anomaly detection. Fabricated in 65nm CMOS, the ADIC has K = 7 Base Learners (BL) with 32 neurons in each BL and dissipates 11.87pJ/OP and 3.35pJ/OP during learning and inference respectively at V dd = 0.75V when all 7 BLs are enabled. Further, evaluated on the NASA bearing dataset, approximately 80% of the chip can be shut down for 99% of the lifetime leading to an energy efficiency of 0.48pJ/OP , an 18.5 times reduction over full-precision computing running at V dd = 1.2V throughout the lifetime.
Spiking Neural Network-based Structural Health Monitoring Hardware System
2021 IEEE Symposium Series on Computational Intelligence (SSCI), 2021
Structural Health Monitoring (SHM) systems estimate damages that affect structural reliability. Modern SHM requires continuous monitoring to detect damage caused by day-today loading and usage. SHM analyzes structural integrity and identifies weaknesses leading to potential building collapse. In disaster situations e.g., earthquakes, SHM systems enable early assessment of building safety and therefore ensure evacuation and prevention of human losses. In an urban area, multi-story constructions in congested areas become a hazard if their structural health is not well monitored and maintained properly. A key challenge is the ability to detect damages in an efficient manner for edge computing. In this work, we propose an SNN based low-cost, energy-efficient, and standalone damage classification model for SHM. The proposed classification model is implemented on FPGAs and results show a classification accuracy of 99.46% on a sensory dataset for earthquake damage on a 7-Story concrete building.
Low-Voltage Energy Efficient Neural Inference by Leveraging Fault Detection Techniques
2021
Operating at reduced voltages offers substantial energy efficiency improvement but at the expense of increasing the probability of computational errors due to hardware faults. In this context, we targeted Deep Neural Networks (DNN) as emerging energy hungry building blocks in embedded applications. Without an error feedback mechanism, blind voltage down-scaling will result in degraded accuracy or total system failure. To enable safe voltage down-scaling, in this paper two solutions based on Self-Supervised Learning (SSL) and Algorithm Based Fault Tolerance (ABFT) were developed. A DNN model trained on MNIST data-set was deployed on a Field Programmable Gate Array (FPGA) that operated at reduced voltages and employed the proposed schemes. The SSL approach provides extremely low-overhead (≈0.2%) fault detection at the cost of lower error coverage and extra training, while ABFT incurs less than 8%overheads at run-time with close to 100% error detection rate. By using the solutions, sub...
MEMS Accelerometers: Testing and Practical Approach for Smart Sensing and Machinery Diagnostics
Microsystems and Nanosystems, 2016
Wireless sensor networks with on-board signal processing capabilities are becoming very attractive for machinery condition monitoring purposes. Their advantages in cost and size are becoming an important factor for deployment in a new generation of maintenance-e-maintenance. In this paper, an intelligent monitoring system for e-maintenance (IMSEM) is presented using a recently developed and tested MEMS accelerometer, a low-power microprocessor and a wireless communication module. The system has a compatible framework and interface with Open Systems Architecture for Condition-Based Maintenance (OSA-CBM). Importantly, it integrates OSA-CBM-defined functions, including sensing module, signal processing, condition monitoring and health assessment; thus, the developed system can successfully reducethe monitoring complexity and communication overhead with human operators. The performance of IMSEM is evaluated by carrying out fault diagnostic on rotating unbalance of a mechanical shaft driven by a DC motor with varying loads and speeds. By analysing the signals from the vibration signal and rotating speed, the IMSEM has the ability to carry out on-board fault detection and diagnosis. The IMSEM achieved an accuracy of 95% in detecting the fault and severity of rotor unbalance, which is based on the ISO 1940-1:2003.