Industrial Solutions

Predictive Maintenance

Application Summary

Predictive maintenance (PdM) systems seek to provide equipment operators and factory maintenance personnel with advance knowledge of impending machine faults. These insights allow proactive intervention to avoid expensive failures and downtime. Designing intelligent industrial sensors for PdM applications is not trivial given the need to address a broad range of end-user equipment installations and operating conditions.

Common Challenges

Cloud based and remote sensor analytics are limited by network performance

Deep learning requires excessive power and memory for edge deployment

Canned algorithms neglect installation specific variations impacting performance

Neural networks require impractically large real-world system data collection

Model complexity and data needs grow exponentially with sources of variation

Involving machine operators in model training desired but not straightforward

SensiML Solution

SensiML Analytics Toolkit has been developed for scalable data collection from sensors ranging from temperature and limit switch inputs through high-rate ultrasonic and acoustic emission sensors. Our toolkit was developed to overcome the practical challenges in predictive maintenance algorithm design and has been tested in numerous enterprise level factory PdM trials and implementations.

Solution Highlights:

Algorithms runs autonomously on sensor endpoints for real-time insight

Draws from broad library of features / classifiers to maximize algorithm efficiency

Generalized algorithms with means for edge learned customization and tuning

Appreciably smaller training data than NNs = practical real-world collection

Collection and labeling workflows that efficiently annotate all variance sources

Means to extend model training for ongoing runtime adaptation by operators

For industrial IoT device manufacturers seeking to build intelligence into their predictive maintenance applications, SensiML offers a uniquely efficient software toolkit to support practical algorithm development that can characterize real-world systems accurately. SensiML’s multiple PdM example applications can be used as a starting point to quickly validate the toolkit's capabilities for specific use cases and rapid proof-of-concept testing.

Anomaly Detection

Application Summary

The most straightforward of approaches for machine PdM monitoring, anomaly detection does not attempt to label various machine fault states but rather just captures baseline operating behavior and normal system variation. By establishing such a reference across one to many sensed inputs, thresholds can be defined for which events triggers can indicate anomalous machine behavior to be investigated.

Common Challenges

Little or no insight into nature of anomalous events

Establishing appropriate alarm thresholds can be trial-and-error process

Potential for many false positive event triggers

SensiML Solution

Using SensiML generated algorithms, developers and integrators of programmable sensor devices can offer industrial end-users rapidly customizable smart sensors. As machine-specific anomaly detectors, sensors can be built with custom trained responses programmed on-site during system commissioning or retrofit.

Solution Highlights:

Baseline algorithms can adapt with edge learning to improve insight with use

Triggered events can optionally report feature vector results or raw data buffer

Simple closed-loop feedback can involve operators in system tuning as appropriate

Using SensiML model tuning and customization capabilities, simple anomaly models can be evolved over time to provide implementation specific learned insights over time.

Process Control and Inspection

Application Summary

High-volume manufacturing processes typically include one or more critical sort / inspection stages that must operate in real-time at high line speeds. These control points use sensor measurements to make decisions directly impacting product yields and overall quality control by comparing measured physical product properties against established reference patterns for classification and real-time action (i.e. sorting, rework, reporting, and upstream process adaptations).

Common Challenges

Cloud based and remote sensor analytics are limited by network performance

Neural networks can require inordinately large training datasets

Hand-coded algorithms are not flexible to accommodate line changes

SensiML Solution

As important as the physical measurement itself, effective sort/inspect systems must possess the ability to analyze the rich sensor signals to detect often subtle differences before they become major excursions leading to rework or scrap. SensiML generated algorithms combined context data and multiple sensors provide the intelligence and real-time responsiveness to support high performance inspection and sort.

Industrial sensor vendors and systems integrators selling solutions to end-user manufacturers can truly differentiate their offerings working with SensiML to build adaptive machine learning into their products and projects.

Industrial Wearables

Application Summary

The application of wearable technology to industrial worker safety and productivity is gaining momentum. The application of sensor driven algorithms as popularized for motion recognition in consumer fitness devices has the potential for far greater ROI when applied in worker safety and productivity.

Common Challenges

User specific motions and gestures requiring user level customization

Power efficient sensor processing as required to achieve long battery life

Cost optimization to achieve competitive pricing for broad adoption

SensiML Solution

SensiML toolkit originated from work in developing simple data driven methods for advanced motion sensing wearable devices. As such, SensiML has many proven examples and sample applications for human motion analysis using low-cost multi-axis inertial sensors and audio.

Examples Include:

First responder wearables for activity detection and gesture recognition

Real-time monitoring of activity and motion

Gesture recognition as augmented communication mode to voice

Gait and postural analysis for worker feedback and injury prevention

Smart tooling to monitor human/machine interactions and safety concerns