Using artificial intelligence at the edge

One of the key benefits of artificial intelligence (AI) is the valuable insight that it delivers to the broader Internet of Things (IoT) strategy. And in no industry is this actionable and insightful data more valuable than in manufacturing.

Without AI, manufacturers have limited visibility into the health and behaviour of their assets (i.e., equipment and devices), and that could have significant impact on performance, cost and security. For AI to reach its potential, each part of the asset value chain needs to be able to gain insight into device behaviour while connectivity costs are controlled.

Original equipment manufacturers (OEMs) face two key challenges with IoT deployments.

First, they have limited insight into the health and performance of their smart equipment/devices once they are deployed into the field, causing them to miss out on some critical information that could help not only avoid unexpected downtime and catastrophic failures, but also deliver better products to market.

Second, when OEMs do gain visibility and access to the data their solutions are producing, the cost of sending that volume of data to the cloud for processing and storage can be exorbitant.

Driving intelligence to the device OEM

Regardless of whether AI-enabled IoT solutions are part of an industrial or consumer implementation, great benefit can be derived if they are added. When device manufacturers embed AI into their devices, they are able to define the inputs and outputs more specifically.

For example, for pipeline monitoring, pressure, volume, and flow rates may be important factors to measure. For industrial equipment, such as robotic arms in a manufacturing line, cycle rates and temperatures may be most significant.

And in consumer goods, such as a dishwasher or washing machine, vibration and energy usage may be the key factors to measure. The device manager can take the AI-enabled platform and feed it into a machine learning engine and monitor the impact on outputs.

When device manufacturers embed AI into their devices, they are able to define the inputs and outputs more specifically.

This process not only brings visibility into the health of a specific device, but also allows the device OEM to gain insights that help health and performance of its entire portfolio. Take household appliances, for example.

An OEM may want the data output to go to their IT systems to run further analysis on their machines. They can use this data to gain intelligence on how a single machine, all machines in a specific model number, or even machines in a geographic location are performing.

Being able to aggregate this data through AI-powered solutions allows OEMs to better understand why things go wrong and determine what factors led to the issue.

For example, was the problem contained to a particular facility? Or to a particular line where parts were made? Are parts failing more frequently? What additional metrics should be measured?

Furthermore, these OEMs can then bring a servitization model into play by offering a maintenance service for an asset that is showing signs of failure, prior to the asset failing.

Deploying machine learning

Translate this to the automotive industry, where AI-powered edge solutions can predict vehicle health based on similar situations with other vehicles. The vehicle manufacturer can be proactive with service announcements, alerting customers that their part may fail and should be serviced to mitigate it.

This data could also be used to invoke warranties, for example, if instructions state that a device is for indoor use, but humidity and conductivity are registering at exceptionally high levels, that could indicate that the device may have been placed outside.

Many OEMs manufacture more than one type of product, and machine learning technology is emerging that allows this level of intelligence and insight to be trained, delivering value back to the OEM without reinventing the wheel. The OEM simply needs to select what data should be gathered as input and output.

This rapidly accelerates time to market to bring embedded AI to devices, because no new training model is needed for each device – it’s just a matter of connecting to the platform.

While different types of devices have different data inputs, the machine learning platform is agnostic to those data inputs. The device simply goes through the training phase to learn normal asset behaviour, and whenever that behaviour deviates from the norm, it triggers an action, such as an alert.

The outcomes of an OEM using edge AI-enabled IoT include increased hardware/asset reliability and productivity, and greater visibility into device performance, enabling product development to plan future improvements.

Controlling costs

One of the biggest expenses of using an IoT solution is the cost of connectivity to transport raw data to the cloud for processing. Most AI solutions don’t take into account the impact on costs when data is sent to the cloud.

Most of that data is reporting readings that are in normal range where no action is needed. Many IoT implementations use cellular or, in some cases, satellite networks for data transmission to the cloud, and there is cost associated with every byte of data transmitted.

When readings are in normal range, moving all data – good, bad and marginal – to the cloud is expensive, and in most instances not necessary. What businesses really want to be alerted to are unusual readings that indicate impending signs of failure, so they can take action before catastrophic operational downtime occurs.

When OEMs do gain visibility and access to the data their solutions are producing, the cost of sending that volume of data to the cloud for processing and storage can be exorbitant.

Technology is emerging that lets all data to be trained, collected and processed at the network edge, with only out-of-range data being transmitted to the cloud. This helps create significant cost savings from both a data transmission standpoint as well as a storage one, as select data can now be parsed and stored for deeper analysis.

A side effect of transmitting all data to the cloud is power consumption. Most IoT devices work on batteries and whenever IoT devices connect to the network and transmit, battery consumption comes into play. By reducing the amount of time devices connect to the network because only select data is transmitted, overall battery life is improved.

Driving new efficiencies

AI-based IoT solutions are driving new efficiencies across a number of industries, and its value to manufacturing is being proven through a variety of use cases. Key to this is controlling costs, and emerging technologies can help analyze data at the network edge, avoiding high cloud transmission costs.

With new visibility provided by AI solutions, OEMs can access the data being collected by IoT devices, glean short-term insights, and also create an environment for longer-term analysis and overall value.

_____

Chris Catterton is the director of solution engineering at ONE Tech.

This article originally appeared in the November/December 2020 issue of Manufacturing AUTOMATION.