About Predictive Maintenance & Anomaly Detection Forum 2023

In photovoltaic grids, charging stations, and home inverters, there exist AC as well as DC systems. Unlike in the case of AC systems, as there is no zero-crossing point in the current waveform for DC systems. When wires get worn off or contacts get broken, the resultant DC arc is more sustainable as a result, it stands to be a greater threat and a huge fire hazard. Series arc fault leads to a reduced fault current resulting in increased impedance and, hence, cannot be detected by traditional protection devices. As a result, extensive research has been conducted on different techniques for series arc detection.

The UL 1699B safety standard provides test procedures to qualify an arc fault detection device. It includes tests for both the effectiveness of detecting as well as immunity to false detection. Frequency-domain-based arc fault detection methods involving spectral analysis by Fourier transform methods are the most popular. Using Fast Fourier Transform makes signal processing computationally efficient, and has been implemented in microcontrollers for household purposes extensively in the current world. However, the spectral signature changes as the device undergoes aging and thereby ends up lowering the accuracy of the conventional algorithm. Since Time-domain methods are computationally most efficient, but conventional algorithms are prone to nuisance tripping, it’s time to use Deep Learning to tackle such problems. 

The data used in this experiment was created with an arc generation setup under laboratory conditions for a variety of voltages (200V, 300V, 400V) and different current (5A,6A,7A) combinations. Raw data was sampled at 313kHz.

The solution presented here is a variable resolution deep learning classification network comprising 4 layers of CNN with a channel depth of 16 each followed by a Fully connected network. The computational complexity of the network is 557k MACs. Multiple CNNs were experimented with to arrive at the simplest and best network to tackle this problem. The dynamic resolution here implies that the time domain input sequence is comprised of 750 data points. This means that the sequence duration chosen for it to be identified as an arc or not varied between 30ms to 250ms whilst simultaneously varying the frequency in the opposite direction. That is, for a 250ms window – data was sampled at 3kHz to effectively give 750 data points, and a 50ms window was sampled at 15kHz to give a 750 data point input. This dynamic scaling ended up boosting the system arc detection accuracy to upwards of 97% compared to the 91% accuracy given by the conventional FFT algorithm. These experiments are tested on a TI 120MHz C2000TM MCU with 69kB on-chip RAM.

The added benefit of this system lies in the ability to retrain this system dynamically on the field which is very much a necessity due to the aging of the system. Given the conventional FFT-based approach, this would be a hard ask as the frequency characteristics change in an unpredictable direction which makes it harder to retrain. Although, the presented work is showcased for arc detection but can be extended for any application involving real-time 1D time series data.

Many motor driven household and commercial appliances play a vital role in our daily lives. Unexpected motor failures in these appliances can result in inconvenience and costly repairs. This presentation explores the application of tiny Machine Learning (tinyML) techniques in predicting and preventing appliance motor failures, without the need for additional microprocessors or sensors in the system and without having to redesign all the motor control electronics already deployed. We will showcase the research from NXP where motor control data is leveraged to determine the health of the motor and detect potential anomalies.

Summary: weeteq’s approach to TinyML, enables circuit-level, sensor-independent, predictive performance planning and unsupervised performance improvement for every closed-loop control system, successfully accessing and optimising circuit-level operational data, from within the control loop for the very first time.

weeteq is pioneering a new technology category of ‘circuit-level machine learning’, accelerating the digital transformation of global industries, ranging from smart manufacturing and industrial automation to domestic appliances and smart homes.  Our technology enables circuit-level, sensor-independent, predictive performance planning and unsupervised performance improvement for every closed-loop control system.

Our presentation focuses on furthering the TinyML approach, to include its application at the micro-controller level, not only providing access to circuit-level operational data, but also enabling real-time in-the-loop analysis and optimisation for the very first time.

We call this technology Ultra Edge®.

Ultra Edge® initial use cases centre on motor and motion control applications within industrial and commercial automation industries. Industries at the forefront of innovation. Innovation in this space, often focuses on deploying sensor, network and cloud infrastructure to better understand and take action within their operating environments.  weeteq challenges this approach, developing circuit-level machine learning solutions which support device-level data analysis and optimisation.

we champion the belief that ‘every machine matters’.

Motor control optimisation results will be presented, which demonstrate how, through the application of TinyML techniques with weeteq’ supplementary control architecture, unsupervised devices can self-optimise, increase productivity and reduce energy consumption.

The TinyML community will have the opportunity to participate in collaboration projects with weeteq and its global technology partners.

 

There are millions of commercial buildings in the world, from offices, to schools, hospitals and stores. Most of these buildings are fitted with heating, ventilation and air conditioning (HVAC) systems, in order to provide a productive and comfortable indoor climate. As with all electromechanical systems, HVAC systems are subject to breakdowns and must be maintained in order to keep them functional. The most common strategy is time-based inspections by the building operations personnel and HVAC service providers. This labor-intensive approach is costly, but many failures still slip by because they happen between inspection intervals.

In early 2022, Soundsensing started to deploy condition monitoring to HVAC systems in commercial buildings. The solution uses vibration- and sound-sensors to monitor rotating equipment in these systems. We find that the system is able to provide early-warning for many mechanical issues, enabling operations teams to schedule maintenance before the system goes down. Furthermore, we find that many systems repeatedly have problems with running when they should not, or not running when they should – and that most of the buildings lack Building Management Systems that accurately detect these issues. Both of these failure modes are monitored using an Anomaly Detection approach, and
combined, the solution provides both cost savings and lower downtime.

Microchip is enabling embedded designers to run predictive maintenance and anomaly detection at the far edge such as 8Bits, 16Bits or 32 Bits MCU/MPU. Engineers as well as data scientist can quickly build their own model with the new MPLab Machine Leaning development suite as it is addressing the full flow of Machine Learning from data collection to deploymen

During this session you will discover how Watteco is revolutionizing the after-market predictive maintenance industry with the help of STMicroelectronics’ technology.
Our team will showcase our adaptable anomaly detection approach that is tailored to fit the unique constraints of after-market solutions. We will also discuss the challenges we faced and how we overcame them to enable optimize and affordable tiny Machine Learning to ease the digitalization of the industry. Don’t miss this opportunity to learn about the latest advancements in predictive maintenance and how you can benefit from them

Railway systems play a vital role in modern transportation and Predictive-Cognitive Maintenance (PCM) has emerged as a transformative approach in the context of Advanced Integrated Railway Management for ensuring the safety, reliability, and efficiency of these systems. PCM leverages data analytics and machine learning to optimize railway system maintenance. This requires effective structural health monitoring (SHM) using low-cost sensor devices. This paper presents a prototype solar-powered wireless sensor node with a 3-axis MEMS accelerometer and energy-harvesting features
for monitoring rail track vibrations. The node contains a microcontroller that runs embedded machine-learning models to preprocess the vibration data after train crossing. Abnormal vibrations, indicative of defects, were detected in real time using the TinyML inference at the edge. Instead of raw data, only the model results were wirelessly transmitted to a digital twin in the cloud. The digital twin aggregates data across the rail network for system-level assessment of the RUL and maintenance planning. This edge
computing approach minimizes wireless transmission and cloud storage compared to raw sensor streaming. Embedded ML enables real-time damage detection, whereas the cloud digital twin enables system-level prognosis insights. The solar-powered platform enables long-term remote monitoring at low cost without wiring or battery changes. A full-scale physical model was used to validate the edge node prototypes against calculation models and wired accelerometers for impulse loads. The results demonstrate that these nodes can provide a sensor layer for cost-effective PCM in railway systems.
In summary, this work proposes an edge computing and embedded ML approach for SHM that integrates cloud-based digital twins to enable the predictive-cognitive maintenance of railway infrastructure. Wireless nodes demonstrate potential for low- cost, convenient, and automated rail health monitoring.

Edge AI is a rapidly developing field with the potential to revolutionize industrial manufacturing. However, successful deployment of edge AI solutions in industrial environments poses unique challenges extend far beyond sensor data and model accuracy. In this talk, we will discuss the practical application of edge AI condition-based monitoring (CbM) in TDK’s industrial manufacturing processes for capacitor packaging.

Capacitor packaging is a rapid process with high throughput and stringent quality requirements. This demands wide-band sensors that can generate data at high rates and fast and accurate AI decision-making at the edge to detect and respond to machine faults in real-time. Where factory automation demands that solutions must be situated amongst a number of existing hardware and software components which control and monitor the manufacturing eco-system.

We will share our experiences and insights on how to overcome the challenges associated with AI development and deployment for industrial manufacturing environments. Our talk will cover:

• The benefits of edge AI CbM for TDK’s capacitor packaging process
• The challenges of deploying edge AI CbM solutions in an industrial environment
• Our approach to developing and deploying edge AI solutions for TDK’s capacitor packaging process
• Lessons we learned along the way

This talk is for anyone interested in the development and practical application of edge AI condition based monitoring solutions for industrial manufacturing.