AI Case Study
MIT Department of Mechanical Engineering developed smart power outlets that distinguish dangerous electrical from benign spikes with 99.95% accuracy with machine learning
MIT’s Department of Mechanical Engineering have designed a 'smart power outlet' to help identify the difference between harmless electrical arcs and dangerous ones for safety reasons. The supervised machine learning model was fed electrical current data, labelled as 'good' or 'bad'. The system monitored the electric flows helping it learn to identify unique devices which posed a potential threat from their currents. The researchers claim that after sufficient training, their 'smart power outlet' was able to identify dangerous electrical arcs with 99.95% accuracy.
Public And Social Sector
Education And Academia
Venture Beat reports that a "team at MIT’s Department of Mechanical Engineering solved the problem with artificial intelligence. Their “smart power outlet,” which they detail in a paper published in the journal Engineering Applications of Artificial Intelligence, analyzes the power draw of appliances and can tell the difference between harmless electrical arcs and dangerous ones. It’s also capable of identifying which devices are plugged in by their unique electrical patterns.
'We create fingerprints of current data, and we’re labeling them as good or bad, or what individual device they are,' Joshua Siegel, a research scientist on the project, told MIT News. 'There are the good fingerprints, and then the fingerprints of the things that burn your house down. Our job in the near-term is to figure out what’s going to burn down your house and what won’t, and in the long-term, figure out exactly what’s plugged in where.'
The innards of the researchers’ smart outlet differ from conventional arc-fault detectors, which contain low-power chips with basic algorithms. A Raspberry Pi sits at the heart of the intelligent outlet, running the machine learning algorithm. An inductive power clamp secures to an outlet’s wire and monitors the current from the accompanying magnetic field, and a USB sound card reads the current data. (Sound cards are designed to detect small signals at high data rates, the team wrote, which make them ideal for this sort of work.)
The researchers fed the neural network power signature data from four different devices: a fan, an iMac, a stovetop burner, and an air purifier. After sufficient training, it was able to identify dangerous electrical arcs with 99.95 percent accuracy (higher than existing AFCIs) and differentiate between the four devices with 95.61 percent accuracy."
From MIT News: "To reduce the risk of fire, modern homes may make use of an arc fault circuit interrupter (AFCI), a device that interrupts faulty circuits when it senses certain potentially dangerous electrical patterns. This is a sensor installed behind the wall that trip an outlet’s circuit when it detects a potentially dangerous spark in the power line. Trouble is, arc-fault detectors are sometimes too sensitive, and, as a result, switch off the power prematurely or unnecessarily."
Researchers claim that after sufficient training, their 'smart power outlet' was able to identify dangerous electrical arcs with 99.95% accuracy, higher than existing AFCIs with the same latency, and differentiate between the four devices it was trained with with 95.61% accuracy.
From the research paper: "We collected data from each device under real-world use to ensure that classification results depend not on signal amplitude or periodic features, but rather on invariant waveform ‘signatures’. We treated classifier development as a supervised learning problem. To test neural networks’ viability on constrained hardware, we implemented a DNN classifier. We implemented a fully-connected DNN in TensorFlow. The model takes as input the generated 1D vector 𝐹𝐴𝐿𝐿, and outputs a probability distribution over predefined classes. In our experiments, we found that a model comprising three hidden layers with the number of neurons being 16, 32 and 16, respectively, works well."
From the research paper: "We instead generated data from an electric stove-top burner, an iMac computer, a fan, and an ozone generator. The burner simulates an ideal resistive circuit, the iMac switching power supply introduces noise, and the fan’s DC motor arcs by design. The ozone generator represents a continuous series fault, as it relies on a high-voltage discharge to cause
an arc between two metallic grid plates, resulting in the continuous formation of 𝑂3