Machine Data¶

This section describes how to integrate machine data with Senseye. Machine data is a broad term that covers all data to be used for predictive maintenance purposes, whether a condition indicator or contextual data.

Senseye includes a cloud data storage solution which can be integrated with via several methods:

• API
• MQTT
• E-mail
• S3 / Azure Blob Storage

Alternatively, Senseye can connect to your historian to collect the data directly.

Currently, we support two types of machine data:

• Time Series Data
• Vibration Data

Time Series Data¶

Senseye supports the ingestion of time series data with a maximum frequency for a single time series (known as a measure) of 1Hz, which can be either a condition indicator or contextual data. Condition indicators, along with relevant contextual data, are used by Senseye's analytics to monitor machines (known as assets) for signs of degradation. For data sampled more frequently than 1Hz preprocessing is required on the data to downsample it so it can be ingested into Senseye.

Vibration Data¶

Senseye supports the ingestion of velocity and acceleration waveforms, and can process these into frequency spectra. Alternatively, if the raw waveform is not available, we can ingest spectra directly. Our system can handle either CSV or JSON in our own formats, or we additionally have built-in support for exports from third-party software.

From raw waveforms, we can extract common condition indicators such as the RMS, Peak-to-Peak and Zero-to-Peak values. When Senseye receives a frequency spectrum or processes a waveform into one, we attempt to extract the amplitude at key frequencies as measures, which are then trackable over time as additional condition indicators of your machinery. Currently, we will extract the amplitude matching the rotational speed of your machine and it's harmonics, which we label as "1x" - "5x". The rotational speed can be sent along with the sample, or alternatively we can allow a fixed speed on a per-sensor basis so that we can perform this calculation. We do not currently support extracting amplitudes at additional key frequencies.

Our method for extracting the amplitude at a given frequency is to take a window of 1% of the target frequency and take the maximum value within this range. So for example, if we had measured amplitudes of 1.1 and 1.2 at 100Hz and 101Hz, and our target frequency was 100.5Hz, we would extract the value of 1.2.