MODD

Motif Discovery and Detection

Time series analysis, automatic pattern discovery and online detection application. Our unique algorithm doesn’t need to know what the pattern looks like, nor how long it is. All we need is data.

Meets your needs in

BigData analysis

Discrete log from huge time series data

Huge amounts of time series data are impossible to analyze manually. A meaningful simplification of data is necessary to gain insight in the observed process. MODD transforms dense time series data into a condensed log of only-meaningful discrete events.

Process mining follows (from discrete log)

Once the data are in form of a discrete log of events, it can be sent to process mining, which extracts dependencies and process description bringing insights otherwise hidden in vast amounts of sensory data.

Quick orientation in data

Categorization

Focus only on stuff that is interesting in data, filter the rest. MODD searches for interesting stuff for you, which jump-starts your insight in data.

Anomalies

There is something else happening at that time than usually? MODD detects it. There is repeatedly happening something new? MODD detects it.

Shape of typical pattern

What does some typical operation look like? What variations does it have? Does it drift in time? Perhaps it changes shape with temperature? MODD prepares views for you to gain insight.

Anomaly detection

Anomaly can be something that happens when not expected, or something that doesn’t when expected. In both cases, a baseline model of what is a typical expected behaviour is necessary. MODD is building such model with the least manual efforts possible.

Correlation analysis

Once the domain-agnostic analysis is done, associations with events coming from outside the data are examined. High-quality product may have been produced with a particular shape of pattern observed on process sensors. Correlation doesn’t imply causality, but it is a first call for a closer look.

AI training datasets preparation

AI is everywhere, but when you need to train your own AI, you need first a lot of data and then a lot of time of experts that prepare the data for AI to be trained on. MODD prepares dataset automatically, highly reducing the time needed for data preparation. After all, it uses this process for training its own detection models.

Causality analysis

When external data are provided, we can put them in context with patterns and analyze, what pattern lead to what event or vice versa.

Shape and timing of patterns (repeating of patterns)

A morning coffee at nine every day is a pattern that doesn’t have to bother you. Sudden overload of server traffic at 2 AM might rise an eyebrow. Does it have the same shape as DB backup, or does it look like an attack?

Pattern drift or evolution detection

When the shape of pattern gradually changes and does not regress back to some nominal form. This can be, e.g., an indication of equipment degradation, or slow change of habits in smart city or smart health applications.

Summarization (compression, precision, deviation)

Create a live concise overview of time series data in form of Gantt diagram once you only need to know when and for how long rather than linger on shapes.

Digital Twin

Patterns form a typical sequence themselves. MODD builds a digital twin that runs in parallel with the monitored process and predicts when will the next pattern happen and which pattern it will likely be.

Features

Domain-agnostic

Whether the data come from smartwatch or from manufacturing robot is not a difference for MODD. MODD looks on the data, and what repeats in it, no matter where the data come from.

Fast

Current discovery algorithms scale quadratically with data, MODD scales linearly with data.

Multichannel

Repeated patterns can be scattered across many sensors and MODD finds them anyway. When detection is run, MODD can sense suspicious deviations in combination of sensors, not only a single sensor. It can detect much more subtle deviations than in single dimension (sensor).

Classification of patterns

Different shapes of patterns can be connected to different events or circumstances. Even if the shapes are very similar to each other, the slight differences can be the difference between a successful and faulty process. MODD learns to differentiate between slight variations of patterns.

Zero-data start (stream)

MODD can be deployed from the start of project and learn the patterns as it goes. Its detections get better with every day of data it collects, as it learns to be better. User can revises MODD models and assigns domain meaning to discovered patterns.

Anytime (early stopping)

Don’t want to wait until MODD examines all the data for repeating patterns? You can query it for its preliminary results any time you wish. Results get better as MODD explores more data.

Cluster analysis

Trivial motifs/patterns suppression

After discovering patterns, MODD examines when the patterns occur and looks for regular occurrences. This way, it suppresses patterns that are just slightly shifted variants of another patterns.

Shape-based clustering

Some motifs are very similar to each other and detection and causality analysis benefits from merging those together. MODD does this with minimal user interaction.

Domains of Application

FinTech

By identifying recurring patterns in daily transactions data, motif discovery provides solid data background for identification of a standard behavior in the transaction network. Having the model of standard behavior enables MODD to raise suspicion when a the model encounters non-standard glitch in the network, i.e., fraud. MODD supervises hundreds of segments of transactions in parallel bringing attention only to the non-standard subset for closer investigation.

Telecommunication

Mobile network communication stations collect great amount of metrics about their usage by customers. When overloaded, an expensive intervention is launched, but this action must also be taken ASAP. MODD allows for online differentiation between a standard and deviated behavior based on highly accurate models for each of the station.

Manufacturing

MODD offers early alerts by monitoring operations for deviations, providing insights into unusual events even in highly repetitive processes. Our system detects anomalies, facilitates timely responses, and enhances overall process efficiency, whether in robotic movements or legacy equipment, such as semiconductor production.

Smart city

MODD optimizes energy consumption by identifying recurring motifs in electricity usage patterns, enabling more efficient energy distribution, cost savings, and reduced environmental impact, among other potential applications.

Smart living

MODD helps smart homes recognize regular patterns in household activities and user preferences. This capability allows homes to automatically adjust energy usage, enhance security, and ensure optimal comfort. MODD also incorporates energy storage solutions like heat accumulators or batteries to intelligently store and release energy based on daily electricity prices and user comfort levels, resulting in a more convenient and energy-efficient living environment.

Smart health

In health monitoring, MODD helps by identifying recurring patterns in biometric data, such as heart rate or sleep patterns. This enables early detection of health issues, personalized recommendations, and improved overall well-being.

Contact us

If you've been inspired by our use cases or have another application in mind where Motif Discovery MODD could make a difference, we'd love to hear from you.