In the next few years, more than 30% of organizations' cloud deployments will include edge computing and edge artificial intelligence (AI) to address bandwidth bottlenecks, reduce latency, and process data for mission-critical decision support in real-time. What's more, 69% of organizations say that prioritizing edge-based analytics will improve their ability to meet IoT objectives for specific use cases.
Indeed, a wide variety of industries, including manufacturing, transportation, and energy, are implementing cloud-edge hybrid strategies to enable advanced, real-time analytics for use cases such as preventive and predictive maintenance, production forecasting, and behavior monitoring. Edge-enabled machine learning (ML) models shift the workload off the cloud and data centers by providing analytics and actionable insights right at the point where data is generated.
Roadblocks to Edge AI Success
When organizations build ML models, an assumption is made that the model will be accurate for a certain time period, as the model has been trained on a particular set of data and accommodates certain data patterns that it is privy to. If new patterns emerge or if the model has not been trained on all possible data sets or workflows, the model might not continue to provide accurate results.
For example, a model can be deployed in a factory to detect defects on a part inspection assembly line or proactively identify patterns that may lead to errors. After a few months, the model's accuracy may diminish due to new data patterns or operating conditions. A degradation in model accuracy can result in significant cost overheads and missed opportunities.
This challenge gave rise to Edge AI capabilities coupled with closed loop ML, which allows a self-directed supervisory mode to continuously evaluate the deployed models for their efficiencies and observe their predictions and outputs when processing variations in data patterns from multiple sensors. This process, which we call closed-loop machine learning, enables the employment of sophisticated machined intelligence and a gating algorithm to analyze and probe deeper into the mind of a machine. Machines equipped with Edge AI functionality can proactively identify mechanical fatigue. This may involve alerting factory operations teams to replace a part, an incorrect calibration with time, or variations in one of the multiple real-world parameters that impact the productive yield of a machine or a process. All of this can be done proactively by analyzing data drifts by an AI model employing closed-loop learning techniques. Learning, re-learning, self-learning aspects of the model is now a reality without the need for more extensive human involvement. With Edge AI, software can proactively interface with live data streams and cater to intelligence at or near the source, leading to increased productivity and efficiency.
Real-world Edge AI Example
Let’s look at a real-life use case where FogHorn's Lightning Edge AI solution made a meaningful impact, saving and improving bottom-line revenue stream for a large industrial manufacturer:
Before FogHorn: If a part being manufactured came out defective, the organization had to deploy several personnel from its IT team, including a big data analyst, a data scientist, an interface developer, and a DB specialist, alongside all actors in its OT spectrum. This process was time-intensive, created orthogonal dependencies, and bred inefficiencies. Also, the ability to experiment with parameter adjustment (say temperature variation), or understanding the causation and result through simulation, proved time-consuming.
With FogHorn: Using intelligent vision-based models and the Lightning Edge AI platform, OT operators receive alerts that highlight errors and aberrations in real-time, as well as its causes, including temperature variation, defective pressure application, etc., periodically. Previously, IT facilitated deployment and periodical maintenance of the system – they were gatekeepers to large-scale deployment in the manufacturing plant. Now, the organization can take remedial measures with a quick turnaround time, as the models are self-learning and will adjust to drifts and inaccuracies.
In summary, using an Edge AI platform, ML models can move beyond traditional analytics capabilities and significantly improve predictive functionality and overall ROI. To learn about FogHorn's edge AI capabilities, check out our product page here, or send us a note here.
Vice President, Product Engineering