Case StudyReducing data footprint with DTECTS

With the availability of our D.Hat PDK, users can now explore the benefits of DTECTS (Detection and Tracking of Energy and Condition Trends System) technologies in their product designs. In this post we discuss how DTECTS KPI Event Vector technology can dramatically reduce data BW and storage requirements while preserving meaning beyond simple level crossing alerts.

While D.Hat implements a conventional three channel power meter adaptable to many different load configurations, including non-energy sensor reading, it also includes an implementation of Ergsense’ DTECTS KPI Event Vector IP that can reduce data footprint dramatically. This technology replaces conventional time series data with compact ‘KPI Event Vectors’ (KEVs) describing changes significant to the user without loss of meaning. These KEVs are based on our expert knowledge of electrical systems and may be customized based on collected system data through manual or machine learning means. Similarly, the user defined KEV trigger levels may be set manually or through machine learning process.

The KEV is useful as an alert and alarm trigger to initiate notification or other processes, and it goes beyond that, carrying useful information regarding the event itself. In the D.Hat PDK FW we have defined the KEVs to deliver data from fast RMS readings of voltage and current in the format:

KEV = {t₀ A₀, t₁ A₁, t₂ A₂, … t_n A_n}, where the terms are in order but not evenly spaced time series data.

t₀ A₀ is the pre event level of the signal
t₁ A₁ is defined as the 10% point of the signal event amplitude
t₂ A₂ is defined as the trigger crossing of the signal event amplitude
t_n-1 A_n-1 is defined as the 90% point of the signal event amplitude
t_n A_n is the post event level of the signal

With the D.Hat FW we have defined a 5 point KEV for both rising and falling events. D.Hat generates a fast RMS signal each approximately every 100us from which filtered and averaged power metrics are assembled into a time series data packet that is sent as fast as every 15s. Watching this time series data can certainly tell the user that events had happened in approximate time, but not exactly when or with what qualities. Using the HW and FW to generate a value crossing trigger can add exact timing to an event, a good step, but still carries no internal information about the event.

To gain deeper meaning from the event, high frequency data must be used. The choice without local intelligence is to send fast time series data to the cloud for analysis at the cost of connectivity and complexity. Sending six channels of data upstream at 10ksps incurs a minimum BW charge of 1.44Mbps, far too costly over cellular networks to be useful on a continuous basis. What is needed is a way to analyze the fast data on the spot and send only the data that is useful in making decisions.

This is where our KEVs come in play. Replacing the time series data normally associated with industrial sensors by a much more compact KEV reduces the transmission of data greatly. Before estimating the savings, we first look at an example of what we can do with the data inside a typical KEV for phase current.

event duration — tracking a change here can indicate unexpected changes in loading or input source resistance
rise (or fall) time — can track this to determine process or bearing loading condition changes
overshoot — for both transformers and motors, an inrush overshoot during startup is normal and not expected to change. Tracking the peak overshoot of an event can tell us if the motor or transformer structure has been compromised.
starting and final current — useful to maintain a watch on process parameters

The examples above demonstrate how a KEV adds value to a simple trigger by allowing analysis of conditions to be done and tracked.

Now let’s look at how using KEVs dramatically reduce BW and storage related costs of sensor data. The following table compares the BW and cost of four cases. Assumed parameters are listed with IoT BW cost being current published data from the internet page of ATT.

Data use / cost comparison graph — Data Use and Cost Comparisons

The KEV at 0.2 cents per month and the KEV + Report, which is a KEV appended with a snapshot of the full D.Hat record of 24 channels, at 10 cents per month than the full time series data at 15s interval provide much more insight than a simple Alert while replacing a time series data cost of $12 per month.

The storage requirements for KEVs and KEV + Reports drop similarly allowing much less reliance on storage at the edge or in your cloud. Machine Learning and AI can operate upon the KEV and Report snapshot data much more efficiently than if crawling through full time series.

All content on this page is an announcement of an engineering evaluation prototype and does not constitute a formal specification or provide for any product performance warranty, fitness for use in any application or imply any regulatory compliance.

DTECTS is Patented (US Patent 10876928) and all its IP is protected under US Patent Law