Predictive maintenance with acoustic sensors

By retrofitting with sensors, industrial companies can make their existing machines and systems fit for modern data analytics solutions. In addition to conventional vibration, temperature and current measurement technology, this also includes new acoustic sensors. This type of sensor is particularly well suited for monitoring the condition of plants and predictive maintenance.

For example, companies can use acoustic sensors to record the noise of critical plant components such as motors, bearings or gearboxes and evaluate it using machine learning algorithms. This enables them to detect unusual noise developments that indicate the impending failure of a component. Through timely maintenance, they can then prevent the machine or plant from coming to a standstill.

There are numerous factors to consider when implementing acoustic sensors. Companies should pay attention to these:

1. Selection. Companies should choose sensors whose frequency ranges are as close as possible to the signal they want to detect. This allows them to avoid costly oversizing, because the larger a sensor's frequency band, the more expensive it is. If the sensors are exposed to moisture or dirt, they should meet the appropriate IP protection classes.
2. Installation. The acoustic sensors should be placed in such a way that they are exposed to as little disturbing ambient noise as possible. If experts can detect anomalies with their human hearing, this is a good indication that the sensors are placed in such a way that the machine analysis will also be successful.
3. Data storage. Companies should check whether they need to continuously record the noise of the monitored component or always have the recording started by a trigger - for example, when the noise exceeds a certain critical decibel limit. Then they can save storage space and costs. They also have the option of storing the audio files on inexpensive media and only storing the metadata in comparatively expensive databases.
4. Data processing. Even if companies do not store the full signal from the sensors, but a processed form, they can save costs. One such form is spectrograms that visualize acoustic signals. They have the added advantage that they can be analyzed with the sophisticated machine learning algorithms that are available today for images.
5. Privacy. Depending on the specific application, acoustic data may contain sensitive information, such as conversations between employees. Companies can filter out such information or prevent its recording from the outset by placing the sensors correctly.

"Acoustic sensors have the potential to add value in many use cases," explains Sebastian Lienert, Data Scientist at LeanBI. "For a successful implementation, companies should follow a holistic concept that includes the selection and placement of the sensor technology as well as the storage, processing and protection of the data. This concept must be tailored to the individual use case. One-size-fits-all solutions do not exist."

Source and further information: www.leanbi.ch​