Edge Computing Meets LoRaWAN Predictive Maintenance Without Cable Clutter

Continuous monitoring of rotating machines often fails due to the immense cabling costs. A new generation of multi-sensors elegantly avoids this problem: the combination of intelligent edge data processing, local ultrasonic analysis and private LoRaWAN networks means that fault detection is shifted directly to the machine.

In the process industry and in traditional discrete manufacturing, countless rotating machines such as pumps, fans, agitators, centrifuges, turbines and compressors are in operation. Efficient and smooth operation of these machines is crucial, as unplanned downtime is extremely expensive. Michael Heider, Head of IIoT Engineering at Wika, explained in his presentation at the rbs trade press days that the absolutely process-critical main units are usually already hardwired into the control technology. The so-called auxiliary units, on the other hand, are often neglected for cost reasons. In practice, condition monitoring here is usually limited to time-controlled, manual rounds by maintenance personnel equipped with handheld measuring devices.

The reason for this glaring automation gap is mainly a financial one: the preparation and cabling costs for installing hundreds of additional sensors, possibly even across potentially explosive ATEX areas, supplying them with power and physically routing them into the company network very quickly exceed the budget for auxiliary units. A completely new architecture is required in order to economically transfer masses of assets to automated condition monitoring: it must be wireless, extremely energy-efficient, offer strong local data pre-processing (edge computing) and, above all, be easy to retrofit during operation.

Gallery

Robust Hardware and Clever Retrofit Installation

Two-part IIoT multi-sensors such as the AsystomSentinel, which measurement technology specialist Wika uses for its condition monitoring solution, are ideal for precisely this demanding requirement profile. In contrast to bulky all-in-one devices, the actual, very compact sensor head is placed directly on the machine - as precisely as possible where the vibration to be monitored occurs. The slightly larger transmitter and computer unit, which also houses the battery, can be mounted remotely using a flat connecting cable. This protects the sensitive radio electronics from strong heat sources or extreme machine vibrations.

A major obstacle in many retrofit projects is the physical fastening: Drilling and tapping on existing motors or pump housings is extremely time-consuming, often requires plant shutdowns and is simply prohibited in many areas. The IIoT sensors, which are specially designed for these purposes and certified in accordance with ATEX directives for hazardous areas, are instead fixed in place using special, high-strength industrial adhesive pads. The fact that this connection technology is absolutely reliable even under the harshest environmental conditions is proven by its use in the harsh offshore sector, where the bonded sensors have been monitoring so-called top drives (power heads) and gigantic mud pumps on drilling platforms for several years.

Edge Intelligence: 50 Parameters and Local FFT Calculation

The real bottleneck in wireless condition monitoring is and remains data transmission. In order to manage hundreds or even thousands of sensors in an industrial wireless network without interference, huge amounts of high-resolution raw data cannot be permanently streamed to the cloud. Signal processing must be relocated as close as possible to the source, i.e. to the edge level.

The microcontroller in the remote sensor module not only records the pure surface temperature of the unit, the ambient temperature and triaxial vibrations via its sensor technology. No, it is also equipped with a highly sensitive acoustic sensor for sound and ultrasound. Around 50 different mathematical parameters are calculated directly on the microcontroller (Edge) from this continuous raw data. These include classic RMS values (Root Mean Square), acceleration, peak values and kurtosis (curvature), which reacts particularly sensitively to shock-like events in rolling bearings. Alternatively, the hardware can also calculate a complete Fast Fourier Transformation (FFT) directly on the chip and then transmit only the finished, highly compressed frequency spectrum to the control system.

The combination of edge computing and ultrasound offers a massive diagnostic advantage over purely conventional vibration measurement: high-frequency vibrations, for example in the 40 kHz range, which are caused by a lack of lubrication or the onset of metal-on-metal friction, can be evaluated locally. They signal impending wear long before the heavy steel housing of the machine even begins to vibrate noticeably and measurably.

High Energy Efficiency For a 10-year Service Life

As the sensors aggregate and evaluate the data locally instead of transmitting it continuously, they are extremely energy-efficient. For rotating machines, it is typically sufficient to send a compact data set every 30 to 60 minutes to monitor the machine's condition closely and reliably. A major motor failure usually builds up over weeks and does not occur every second with this industrial equipment.

In combination with intelligent sleep modes ("wake-up on event"), in which the sensor only wakes up in the event of significant changes, the IIoT nodes achieve operating times of up to ten years. A decisive advantage for subsequent maintenance: instead of permanently installed, proprietary special rechargeable batteries, the Wika system uses four standard AA lithium cells. After simply loosening four screws on the housing, these can be easily and, above all, cost-effectively replaced directly on the machine anywhere in the world.

Private LoRaWAN and Seamless IT/OT Integration

The open wireless standard LoRaWAN (Long Range Wide Area Network) is used for data transfer from the edge device to the central gateway. This energy-efficient LPWAN technology offers massive architectural advantages in harsh industrial environments:

• The system impresses with its long range and penetration. Data can be reliably transmitted over 5 to 10 kilometers (approx. 3.1 to 6.2 mi) to a single gateway, even in environments with heavy metal structures in the process industry. The gateways themselves can be installed in just a few minutes and only require power and a connection via Ethernet or LTE mobile radio.
• Users benefit from a completely independent infrastructure. They build their own private LoRaWAN network and are therefore completely independent of public telecommunications infrastructures or expensive 5G campus licenses.
• The system supports genuine bidirectional communication. Sensors not only send their data packets (uplink), but can also receive configuration commands (downlink), for example to temporarily increase the measuring intervals "over-the-air" if a machine behaves conspicuously.

On the IT side, the gateways send their encrypted packets to a central network server (e.g. from LORIOT). From there, the decoded data flows into the application server, which can run either on-premises in the company's own data center or in the cloud. The processed diagnostic data can be seamlessly integrated into existing ERP or SCADA systems via standard interfaces (such as REST-API or MQTT). This IT-centric approach also enables globally distributed teams: gateways can be installed at a production site in France, while a central vibration expert at the headquarters in Germany monitors the data volumes in real time and provides precise instructions to the maintenance team on site.

Machine Learning: From Raw Value to "Anomaly Score"

However, simply recording all these parameters does not solve the fundamental problem of data overload for maintenance staff. Instead of setting rigid, error-prone threshold values for hundreds of different motors, the integrated software solution learns the exact normal behavior of each individual measuring point in an initial training phase. This phase usually lasts two to three weeks, during which the machine runs in its normal operating conditions.

During operation, the machine learning algorithm then constantly compares the newly arriving parameters with this individually trained reference model. The result is a user-friendly anomaly score from 0 to 100 percent that visualizes deviations at a glance. However, the system goes a decisive step further: the AI uses static machine data (a so-called identity card with values such as power and speed), classifies the deviating symptoms in the spectrum and calculates a percentage similarity score. It automatically identifies probable sources of error - such as gearbox wear (67 percent probability) or loose rotor bars (36 percent). This gives technicians a direct, actionable diagnosis instead of just raw frequency graphs.

Practical Example 1: Ultrasonic Analysis Prevents Reactor Failure

How massively superior this automated AI approach is to manual control was impressively demonstrated at an Italian plastics manufacturer. The Wika solution was installed on a large stirred tank reactor. Sensors were located at the motor output, at the rear of the gearbox and on the drive shaft directly above the mechanical seal to ensure seamless monitoring of the power transmission.

As early as May 2025, the software recorded a steady increase in the anomaly score due to significantly increased ultrasonic levels in the 40 kHz range. A technician then checked the machine with a classic, hand-held vibration measuring device, found no noticeable vibrations and incorrectly classified the message as a false alarm. However, the AI was right: in mid-June 2025, the now advanced wear was reflected in massive, low-frequency housing vibrations shortly before the rolling elements of the ball bearing in the reactor broke completely. It was only thanks to the very early ultrasonic warning that the plant operator was able to plan the timely replacement of components for the next regular plant shutdown and avoid an expensive, potentially dangerous unplanned shutdown.

Practical Example 2: Mining Rollout Delivers Measurable ROI

A global mining group introduced the system in parallel at seven pilot sites to monitor heavy-duty equipment such as large ventilation systems, cooling pumps, winches, vibrating screens and huge rock crushers. Previously, maintenance at these sites was based on reactive measures or fixed maintenance intervals, which often led to unnecessary inspections without any findings.

The radio-based condition monitoring system delivered measurable improvements throughout the entire process chain within just a few months:

• Reduction in downtime: Unplanned downtime at the absolutely critical crushers and mills fell by a significant 40 to 60 percent across all sites.
• Cost savings: The Group saves an estimated 100,000 to 300,000 euros per month (approx. USD 118,000 to 354,000 per month) at the seven pilot sites thanks to the avoided plant downtimes and the higher OEE.
• Return on Experience (REX): The so-called "conversion rate" for maintenance tickets tripled. The mechanics and electricians are now only called out in the event of genuine anomalies that have been confirmed in advance by the AI. This minimizes frustrating false alarms and massively increases acceptance of the digital system among the workforce on site.

The Group easily achieved its commercial target of a return on investment (ROI) of less than six months and is now using the architecture as a blueprint for global scaling.

Scalability Through Edge Intelligence and Radio

The clever combination of local edge data processing and energy-efficient LoRaWAN radio solves the fundamental scaling problem of condition monitoring in existing systems. Instead of extremely expensive and rigid hardwiring, predictive maintenance can be set up quickly, flexibly and budget-friendly via wireless retrofit. Plant operators reduce unplanned downtimes and at the same time relieve their specialist staff of routine manual checks. When the intelligent sensor AI presents the most likely cause of the fault directly to the maintenance staff on the screen, the concept of predictive maintenance becomes a measurable and highly profitable reality on the store floor.(mc)