Abstract
Most of us are familiar with the type of sounds our household devices emit when working properly. In a majority of cases, we can diagnose whether a machine is malfunctioning simply by listening to the sound it is producing. It is as if we familiarize our memory with the sound signature of a device, which we term as normal. We treat any sudden and major departure from the memorized sound as a potential malfunction. This is why we apply lubricating oil to the hinge of a squeaking door or window, and we stop the car and pop its hood to investigate a knocking from the engine.
Acoustic sensing and diagnosis in the factory
In the industry too, moving machinery produces a characteristic sound pattern that most operators can certify whether coming from a healthy operation. According to Scully, 1 experienced technicians can often sense and identify changes in machine behavior simply by listening, not requiring instruments for the process. This is acoustic sensing and diagnosing using sound waves through air and is different from temperature and vibration sensing, as it does not use data from numerous sensors placed at specific diagnostic sites. Technicians usually develop such skills after many years on the job.
The industry has been using airborne ultrasonic sensing as a tool for machine diagnostics and predictive maintenance for several decades. Changes in machine behavior from congestion, damaged bearings, and other issues cause small changes in airborne frequencies as low as 40 kHz, which handheld instruments can measure and technicians can use for diagnosis.
According to Scully, 1 a team of engineers and computer scientists is now developing deep learning neural network (DLNN) that they can train to acquire the same skills as the technicians do in acoustic sensing and diagnosis. The team has backgrounds in machine learning for machine vision, and they use deep learning on the cloud for conducting sound-based diagnosing for different components in a factory.
Using ultrasonic microphones placed near machines, they record its sound continually, while sending the audio data to a cloud-connected unit. Here, DLNNs process the data and classify the sounds as healthy operations or any other, as the team’s algorithm specialists diagnose them. The process is quick and the technicians on the floor can attend to problems immediately.
Acoustic sensing and diagnosis in healthcare
The medical fraternity has been using this technique of acoustic sensing and diagnostics for long. They use ultrasound diagnostic devices for various purposes from obstetrics, pediatrics, internal medicines, and intraoperative procedures. Doctors and technicians use this medical device to examine a wide range of patients from children, expectant and nursing mothers, disabled, and elderly persons.
Hitachi Ltd 2 has developed an ultrasonic diagnostic device with a monitor. When a doctor uses it to examine an expectant mother, both can see the image of the fetus safely and easily. The probe has a contour to fit the patient, and the doctor uses it with a gentle low pressure. When the doctor uses the device for internal medicines and pediatrics, the external appearance of the probe allows children and their parents to feel safe and at ease. However, this product is a medical device, for use by medical professionals, and is not for sale to the public.
Acoustic sensing for intelligent gearbox diagnosis
Gao et al. 3 have studied the use of acoustic sensing for intelligent gearbox diagnosis, where it is difficult to obtain information and a large enough sample size to study. They propose various methods for diagnosing gearbox faults, which includes using wavelet lifting, a support vector machine, and rule-based reasoning. Since not all machines may have the same fault, and it is possible that fault features will vary, Gao et al. 3 use a support vector machine for the initial diagnosis. They process the gearbox vibration signals with wavelet packet decomposition, extracting the signal energy coefficients of each frequency band and using them as input feature vectors in the support vector machine for recognizing the normal and faulty patterns.
Next, they use wavelet lifting to filter out the noise signals successfully, while maintaining the impulse characteristics of the fault for precision analysis, thereby effectively extracting the fault frequency of the machine.
Finally, they built the knowledge base based on the field rules summarized by experts. This helped them to identify the detailed fault type. The study concludes that the support vector machine can be a powerful tool for detecting gearbox fault patterns even for small sample sizes, the wavelet lifting scheme effectively extracts fault features, and the rule-based reasoning can identify detailed fault types. Therefore, the combination of support vector machine, wavelet lifting, and rule-based reasoning constitute an effective technique for gearbox fault diagnosis.
Diagnosing and repairing noisy brakes
Noisy brakes can be very annoying, and the loud squeal, grind, squeak, and other noises from the brakes can lead to decisions related to their replacement or repair. Tomashek 4 has compiled a number of reasons for brakes being faulty, and the type of noise they make.
According to Tomashek, 4 noisy brakes can be due to brakes wearing out, brake pad material being low, dirty, improper brake pads, lack of lubrication, and brakes getting hot. However, it is very important to identify the corner of the vehicle making the most noise—Robert suggests taking help of an assistant for doing this. He also gives detailed steps necessary to repair or replace the brakes as necessary.
Using sound as a weapon
According to Science Correspondent Devlin, 5 several people, since the Second World War, have tried to use sound as a weapon against their enemies. Most notable of them was the effort made by the German military, who considered deploying an infrasound source for knocking enemy bombers out of the sky using a vortex of sound. However, low frequencies are difficult to focus and target. Infrasound can be dangerous to people, as it can resonate with their stomach cavity, and cause them to feel sudden nausea or anxiety.
Devlin 5 also discusses the US State Department claims in autumn 2016 that US diplomats based in Cuba were subject to an acoustic attack, which affected at least 16 individuals inflicting on them symptoms such as hearing loss, headaches, and loss of balance. The US State Department claimed that sophisticated devices operating outside the range of audible sounds were responsible for the sonic attacks. However, they were unable to discover any such device or perpetrator.
According to Devlin, 5 Tim Leighton, a professor of ultrasonic and underwater acoustics at the University of Southampton, agrees that pointing a tight beam of energy requires ultrasonic energy. Additionally, evidence from industry points to hearing loss suffered by people exposed to ultrasound used for welding plastic parts. Moreover, building a transmitter for ultrasound is not difficult, as transducers are readily available for higher frequencies above 20 kHz, and anyone with an engineering background could assemble the device.
As Devlin 5 explains, ultrasonic energy does not travel very far, and when faced with barriers such as human skin, curtains, and walls, is highly attenuated. Therefore, to function as an effective weapon affecting human hearing at a distance, the device would have to be powerful enough, requiring a large amplifier, a focused beam, and would necessarily have to be in close vicinity of the target.
According to Devlin 5 and Tim Leighton, a device the size of a kitchen matchbox could emit high enough amplitudes at close range to induce feelings of anxiety or difficulty in concentrating. An ultrasonic beam capable of going through a window would require a lot of power, and the device producing it would start to look more like a suitcase. For generating hearing loss at a distance of 50 m, the device would be about the size of a car. Therefore, Devlin 5 is skeptic about the sonic weapon theory and suggests the use of drugs or poison instead.