Industrial Internet Of Things And Machine Learning Shaping The Future Of Predictive Maintenance

In factories, production facilities and other industrial environments, millions of machinery and equipment help to produce all the items required by people from food and drugs to cars and computers.However, even the best machine cannot run forever, there will always be various problems.When the internal faults such as bearing wear and the heat of the motor, or external conditions such as humidity and temperature cause these key equipment assets such as failure, even a short time failure will have a serious impact on the enterprise.

For example, Siemens artificial intelligence machine health management platform Senseye Predictive Maintenance existA report released in 2022 pointed out that a large factory lost 25 hours of production time due to accidents due to the accident of machine accidents.More than 2 million US dollars.Deloitte Consulting said that the loss of non -planned machine stops to industrial manufacturers each year is estimated to be as high as $ 50 billion [2].

These data show that it is not feasible to wait for the device or machine to be repaired after failure; however, it is not possible to blindly perform maintenance or replacement without reason.Predictive maintenance provides the answer for this.

Prevent machine failure

Predictive maintenance(Traditionally called status monitoring) It is not a new concept. At least it can be traced back to the early 1990s. Its goal is to launch machine maintenance before the equipment performance becomes unsatisfactory when the equipment performance becomes unsatisfactory.In practice, alarm will be issued when the health status of the machine reaches the specified threshold. The engineer carefully checks the condition of the machine, finds defects, and fixes it before the problem deteriorates.

This method has huge value.Deloitte said that the implementation of predictive maintenance can save the cost of materials per year5%to 10%; the normal operation time and availability of the device can be increased by 10%to 20%; the overall maintenance cost can be reduced by 5%to 10%; the maintenance plan can be reduced by 20%to 50%.According to the McKinsey Global Institute of Global Research, predictive maintenance based on the Industrial Internet of Things (IIOT) usually reduces the time of machine shutdown by up to 50%and the life span of the machine can be extended by 40%[3].

The advantages brought by predictive maintenance are not as good as the manufacturing industry, and almost all industries that depend on tangible assets to serve or produce can benefit.For example, power companies can use predictive maintenance and monitoring tools to prevent power outage and avoid huge losses and confusion.

Industrial Internet of Things promotes predictive maintenance

During the early development stage of predictive maintenance, due to the lack of appropriate sensors to collect data, and the computing resources of sorting and analyzing information are limited, it is difficult to implement this maintenance.Today, due toThe advancement of IIOT, cloud computing, data analysis and machine learning (ML) technology, pre -pre -pre -pre -pre -pre -Measurement maintenance in small and medium -sized enterprises(SME) and large enterprises are common.In the Siemens report, about three -quarters of the respondents regard predictive maintenance as a strategic focus.

Low -power Bluetooth and other powerful low -power wireless technologies enable sensors to collect data in networks composed of hundreds of, thousands, or tens of thousands of devices.TheseIIOT sensors monitor parameters such as temperature and vibration, pressure, gas levels and energy consumption, which can make the service team more deeply predict the future state of the equipment, and actively respond before the problem appears to prevent problems before they occur.

DeloicPredicted maintenance positioning files states that data is fuel for any predictive maintenance engine.The quality and quantity of data are limited factors that analyze the root cause and predict the fault in advance.IIOT can provide a steady stream of ' fuel '.The sensor continuously collects data, and then the key information is passed back to the central server or cloud through the gateway.

Wireless networks also have other advantages, such as realizing asset monitoring in the dangerous environment, not recommended for people to go to and difficult to reach.The wireless network can also greatly reduce costs related to the installation, maintenance and performance of predictive maintenance systems.

Machine learning

It is one thing to build a wireless network that can generate millions of data points every day. How to understand all data in time and respond to the needs of the problemThat's another matter.In order to limit the cost and energy consumption of transmission of large amounts of data(Most of the data are mediocre), and people complete most of the calculation work through the technology called edge processing.This requires a lot of computing power and memory, sensor fusion, and more and more machine learning (ML).

Sensor fusion is a combination of sensor data, so that the information obtained is higher than the information from separate sensors.(That is, more accurate or more complete).ML is an application of artificial intelligence (AI), allowing computers to be directly compiled directlyStudy under the condition of Cheng or instruction.The ML algorithm learns from the data and then infer the data that has not been seen, so as to make decisions without a clear instruction.As a result, the ability and high autonomy of machines are constantly enhanced.

Edge processing and sensor fusion enable IoT devices to scree data local screening to determine which data of ordinary situations and which indicate that the situation is changing, and it should be marked.AddML, marginal devices can not only check whether the data exceeds the preset threshold, but also infer what these changes mean these changes, and then take corresponding measures.

A predictive maintenance -related example is to monitor the temperature sensor of the bearing of the machine.This sensor is availableThe ML model is inferred that, for example, the gradual increase of the bearing temperature is just a preheating machine without worrying; but if the bearing temperature rises rapidly, it may indicate that the lubricating failure may occur and the sensor is triggered to cut off the machine before mechanical damage.

Help analytical prediction maintenance

Many advanced todayML models require a lot of computing resources and high energy consumption to perform reasoning.However, although a large number of IoT connection devices today can perform some marginal computing and sensor fusion, these resources cannot be obtained.

Micro -machine learning(Tiny Machine Learning) or Tinyml is a solution.This technology is a branch of ML, but simplified the software, so that the battery -based power embedded device (such as wireless system -level chip (SOC)) based on microcontroller can also be transportedMovement of machine learning.

Nordic design partner Edge Impulse provides Tinyml software, which can run on Nordic's NRF52840, NRF5340 and the latest NRF54H20 SOC.Nordic provides applications that can be used for training and deployment of embedded ML models on the THINGY: 53 IoT prototype platform.This application allows developers to upload the original sensor data to the cloud -based Edge Impulse Studio via mobile device, and deploy the training ML model to Nordic Thingy: 53 through low power consumption Bluetooth.

Nordic IoT prototype platform can be used to test the ML model

This function promotes the development of sensors, such as adoptingNordic NRF52840 SOC's Atomation Atom.This sensor can measure oscillation to determine whether the machine motor is more vibrated than yesterday, or monitor the temperature to check whether the bearing is fever when the machine is running.Each ATOM can work with 3.6 V lithium -ion batteries for up to three years.

Depend onNordic's NRF52840 SOC -driven Atomation Atom can monitor vibration to master the health of the machine

These sensors monitor and process information locally rather than sending constant data flows to the central system.When the threshold or equipment runs exceeding the normal parameter range, the data will be sent to the gateway through a low -power Bluetooth wireless connection.For example,Can the atom device be sure and answer " The device is turned on or closed?" " Is the motor vibration larger than yesterday?" or " Is the bearing temperature too high when runtime?" and other issues.

Atomation CEO Steve Hassell explained: Nordic SOC is the real brain of our Atom product. This brain must receive original sensor data, convert it to useful information, and make independent decisions before communicating in a harsh RF environment.Essence

What is the next step?

Predictive maintenance is more and more usefulML supports automatic analysis technology.This will mean reducing artificial participation and gaining better results, but this also increases the demand for marginal computing power.Nordic has foreseen this market demand, so it has launched the next generation of short -range SOC products NRF54 series.

New typeSOC integrates multiple ARM Cortex-M33 processors and RISC-V processors, and each processor optimizes a specific type of workload.By combined with embedded large -capacity non -easy -to -miss memory and RAM, the NRF54H20 SOC provides developers with more computing and memory resources required to run complex ML -driven predictive maintenance applications.Even better, combined with the use of NRF54H20 and Nordic NPM1300 and other power management IC (PMIC), it can achieve ultra -low power consumption, extend battery life, reduce maintenance demand and reduce waste.

Nordic NRF54H20 SOC supports Tinyml, bringing powerful computing power and memory resources to achieve predictive maintenance applications

Future, byA predictive maintenance solution supported by a new generation of SOC support such as NRF54H20 SOC will be more flexible, efficient and sustainable.This will promote the potential of all types of machine operators to give full play to the potential of assets and extend the operating life.In just thirty years, predictive maintenance has made great progress