Our products > Other Products ​> Advanced Time Series Analytics

Advanced Time Series Analytics

Automated Zero-Day Anomaly Detection and Root Cause Analysis for Sensor Data


As civil and military platforms increase in capability, the number of diverse and interconnected sensors within them, also increases. During testing and evaluation, as well as ongoing operations and maintenance, thousands of sensors generate huge quantities of time-series data. While these data contain valuable insights and could provide visibility into platform behavior and failure mechanisms, their use is often severely limited for practical reasons. 

With the increasing complexity of missions and platforms and the advent of multi-domain operations (MDO), these analytic limitations are a significant and growing concern. Without the ability to identify zero-day anomalies, uncover unanticipated conditions, and pinpoint root causes, critical platforms are subject to unknown vulnerabilities.

​ATTENDS is designed to meet this challenge by providing Advanced Multi-Variate Time Series Analytic Techniques, that can be practically applied to large sets of diverse sensor data. ATTENDS enables and automates knowledge discovery and causal analysis on massive, heterogeneous datasets via novel applications of AI and machine learning (ML) algorithms.
Anchor Element
Copy for linking on the same page:
https://www.peratonlabs.com/?data-scroll-to-anchor=HowItWorks
Copy for linking from an external page:
https://www.peratonlabs.com/?data-anchor-link=HowItWorks

How it works

 ATTENDS uses a streamlined workflow engine to orchestrate data processing for flow-through analytics and to coordinate across data ingestion, storage, training, analysis, execution of AI/ML algorithms, and presentation of results to the user. The workflow manager schedules data ingestion from external sources, necessary pre-processing of data, and training of AI/ ML models.  

The knowledge management module is implemented via directed graphs and contains domain knowledge about the AI/ ML algorithms, datasets, and metadata. Salient knowledge is automatically extracted from the metadata via sophisticated unsupervised learning algorithms. The workflow manager will, as needed, retrieve information from the knowledge store about a dataset to determine how it should be preprocessed to support the training of AI/ML models, including the use of data augmentation if needed.

The ATTENDS architecture is designed not only to support a wide variety of AI/ML tools and algorithms, but also to easily incorporate new AI/ML algorithms with minimal, incremental effort. This is achieved through the AI/ML APIs that support user configurability and remote invocation, and the inclusion of multiple ML learning methods (supervised, semi-supervised, unsupervised, and reinforcement learning).

​ATTENDS has applications for predictive maintenance, including failure detection and remaining useful life (RUL), as well as target location error analysis, zero-day anomaly detection, identification of causality, and root cause analysis. Other analytics applications can be easily added for use cases involving very large datasets and multi-modal sensor data.
Anchor Element
Copy for linking on the same page:
https://www.peratonlabs.com?data-scroll-to-anchor=features
Copy for linking from an external page:
https://www.peratonlabs.com?data-anchor-link=features

Features

  • Fast, automated data ingestion and fusion with a flexible data architecture that supports structured, semi-structured, and unstructured datasets
  • Knowledge management architecture with AI-aware knowledge model to extract information from meta data with minimal human involvement
  • Flexible workflow engine with batch and query modes to orchestrate data processing flows
  • Innovative multi-variate time series analytic methods to capture temporal dynamics in sensor data
  • Automatic configuration, training, and application for a wide range of ML algorithms and training techniques
  • Synthetic data generation to augment training data
  • Unsupervised learning for zero-day anomaly detection and knowledge extraction
  • APIs (application programming interface) to support user configurability and remote invocation
  • Easy-to-use interface to support of test, evaluation, and maintenance operations
Anchor Element
Copy for linking on the same page:
https://www.peratonlabs.com/?data-scroll-to-anchor=Advantages
Copy for linking from an external page:
https://www.peratonlabs.com/?data-anchor-link=Advantages

Advantages

  • Enables data mining and knowledge extraction from complex, voluminous datasets, which had been largely inaccessible due to their massive size
  • Meets mission needs for operational timeliness via fast ingestion and automation expedited anomaly detection
  • Provides time saving orchestration across the data processing and analysis workflow
  • Lightens operator’s workload via sophisticated techniques in time series analytics and AI/ML algorithms and tools
  • Harnesses unsupervised learning to discover “unknown unknowns” and valuable insights that were hidden
  • Deep dive capabilities and causal analysis provide guided problem diagnosis – diagnosis is historically among the hardest problems, requiring very experienced personnel who are scarce and in high demand
  • Provides automated data augmentation as needed for training ML models
  • Proven effectiveness in zero-day anomaly detection, root cause analysis, and target location error detection and analysis
  • Flexible and extensible platform that easily adds new AI/ML algorithms and new analytics applications to support use cases for MDO
Anchor Element
Copy for linking on the same page:
https://www.peratonlabs.com?data-scroll-to-anchor=resources
Copy for linking from an external page:
https://www.peratonlabs.com?data-anchor-link=resources

Resources

Research and products of interest

Research

Anchor Element
Copy for linking on the same page:
https://www.peratonlabs.com?data-scroll-to-anchor=requestinfo
Copy for linking from an external page:
https://www.peratonlabs.com?data-anchor-link=requestinfo