Major Breakthrough by CEMIT - Precision of Digital Twin Technology Proven, Taking Predictive Railway Maintenance to the Next Level

Last week, CEMIT conducted a calibration test that compared their Digital Twin on regular trains,  which is based on data collected from CEMIT IMU Sensors, with manual calibration using highly accurate measuring tools. The results were nearly identical, proving the predictability of the technology and its relevance on an international scale.


Using “Digital Twin” Technology to Preemptively Predict Changes in Track Geometry


Over the past few years, CEMIT has adopted a digital twin design conceptapplying it to the condition-based monitoring of railways. Although digital twin is currently being used within various industries, it is still relatively unused in the rail industry. This empty space is actually surprising, because digital twin has the power to make maintenance more efficient – one of rail’s largest hurdles.


Digital twin is a tool used to meet the new realities of software-driven products. It is a technological leap, if you will, into the core of a physical asset. In essence, it is “a virtual representation of an object or system that spans its lifecycle, is updated from real-time data, and uses simulation, machine learning, and reasoning to help decision-making.” In essence, it’s a detailed, virtual model of a physical thing that offers valuable insight into how that thing is operating in the real world. (IBM Business Operations 2020)


Digital twin, in the context of rail, helps engineers and operators understand the current status of trains and rail infrastructure. It also helps with predicting and how trains will come to perform in the future. For CEMIT, we designed our digital twin solution to exploit the immense potential of the large datasets collected by our IMU sensors.


The Need for Intelligent Monitoring Solutions


As railway infrastructure investments and the market for maintenance machinery continues to grow, we are seeing a parallel demand for intelligent monitoring solutions. Solutions that are low-cost, can detect a wide range of faults before they become serious, and display alerts in real-time in an intuitive and decision-supporting way.


However, most solutions found in today’s market for monitoring and identifying faults are based on hands-on, periodic inspections. While some tech-driven solutions are available, they are costly, tend to be hyper-focused on a single parameter, and are application-specific without the capability to carry out root-cause analysis. Thus, many companies can only apply the solution periodically or across a small percentage of their fleet or infrastructure. As a result, their ability to proactively maintain their assets needs to follow a manual schedule, making them vulnerable to significant fault occurrences that compromise efficiency and safety.


CEMIT, dedicated to making railways more intelligent and building smarter and more sustainable transportation infrastructure, doesn’t see this type of solution as either viable or long-term. Breaking away from the pack, we have chosen to differentiate ourselves from other available technologies by focusing on the use of artificial intelligence to collect huge amounts of data through IMU sensors . These sensors are not only designed to be low-cost, but to be used on regular trains during normal traffic. What’s more is that they are fully compatible with all existing trains, fast and easy to install, and are maintained remotely.


CEMIT’s Digital Twin Based on Large Data Collected From IMU Sensors


CEMIT’s IMU sensors consist of two main parts. The first is called a triaxial accelerometer, which uses three separate sensing elements (oriented at right angles with respect to one another) to analyze vibrations. The second element is called a triaxial gyroscope, which measures rotational motion and angular velocity. Comparatively and in sum, the gyroscope helps indicate orientation through rotation, whereas the accelerometer measures linear acceleration based on vibration. To ensure that the gyroscope and accelerometer measurements are collected onboard and continuously, the IMU sensors are also mounted with GPS antennas.


Once installed, recordings from the IMU sensors flow to CEMIT’s cloud platform. Then they are collected, stored, and processed through the CEMIT AI (Artificial Intelligence) engine. Here, advanced algorithms calculate insights regarding the current state of the track from which they came, enabling the detection of early phase deviations and root-cause errors before they turn into major issues. This is all done with millimeter precision. 


The AI engine also learns by picking up on patterns within the large datasets to predict the evolution of the train’s geometric parameters, as well as predict the probability of deviations within these expected parameters. From this, the algorithms build trends, and these trends help CEMIT’s scientists and engineers pinpoint discrepancies and exact error locations, both in the present and in the future. The CEMIT Digital Twin is built from these data sets, and can both estimate a highly accurate current state, as well as make highly probable predictions of the future development of the tracks.

“The test results of CEMIT’s digital twin and CEMIT’s algorithm for estimating track parameters have proven to be very effective, marking the technology as ‘ready’ for use as one of the leading tools for modern condition monitoring in the railway sector.”

Successful Test Results Proving Millimeter Precision


CEMIT wanted to put the accuracy of our predictive digital twin system to the test by comparing it with real-world measurements. To do this, we first installed a CEMIT IMU sensor on a train that runs daily on a line between Porsgrunn and Brevik. The sensor collected specific data to identify the track’s geometric parameters across different track conditions, including straight and curved sections. This data was then run through our AI engine to build a digital twin with three-dimensional  track geometry.


Next, we needed to calibrate the estimated track data of the digital twin. To do this, we went out to inspect the track and make manual measurements with a railway measuring car, together with a railway engineer. After cross-referencing and comparing our calculated estimates with the manual measurements, CEMIT was pleased to find that our algorithm produced estimates that consistently reflected the manual measurements, within just millimeters of one another.


The figure above shows the estimated cant (h – measured in millimeters) – on a straight stretch of track as a function of mileage (s – measured in meters). The red curve represents estimated cant, and the black curve represents manually measured data. The results are compared with a nominal cant in blue, equal to zero along the straight track line.


As you can see, CEMIT’s algorithm was able to identify the areas of the track with the wrong cant, according to the nominal value. The minor discrepancies between the estimated and measured cant arose because the predicted estimates were based on IMU Sensors, which reflected train-on-track measurements. In contrast, the manual measurements were made on bare tracks without the load of the trains.




With this calibration, we believe to have successfully proven CEMIT’s Digital Twin’s ability to produce accurate cant profiles that systematically monitor the development of errors. Our alerts inform maintenance teams ahead of time when a fault will hit a critical level, as well as immediate acute errors, so that maintenance can be planned accordingly. These automated alerts can also be set in advance for areas of possible concern, allowing teams to track their progression.


In summary, CEMIT’s technology offers an innovative, cost-effective, predictive monitoring platform that reduces failure rates and improves operational efficiency for the railway industry. With measurements linked to particular positions on the track, catenary, or train, each reflected in alerts that provide accurate geographical positioning, CEMIT technology directly leads to fewer misdiagnoses that save money, time, and operational headaches.