top of page

Mobbot IoT and Data Analysis Solutions for Tunneling

Contractors in the tunneling industry use a wide variety of machines for different applications as well as different measuring devices to capture physical phenomena. What is often missing is a clear and easy-to-read overview of all measured data and the possibility of cross-comparison for a better understanding of the process. For sprayed concrete applications, this could be achieved with digital data collection tools, such as the telemetry and data analysis solutions from the Swiss company Mobbot. By bundling data, several project participants could benefit from its use despite different requirements.

For example, what happens on the job site will most likely be clear to the operators, but not necessarily to the same extent to site managers. Site managers and remote teams base their understanding on both complex handwritten reports and numerous visits to the tunnel. If such information is made available digitally and online instead, a lot of time can be saved. Moreover, the online availability of consistent information from multiple devices and reports allows site managers to keep track of materials used and optimize their consumption. This not only reduces the carbon footprint of the construction project – it also offers opportunities for cost savings.

By monitoring the historical values on machine behavior over a longer period of time, it would be possible to perform machine maintenance and replace parts more quickly. Furthermore, valuable information about recurring machine failures can help manufacturers to systematically optimize their products when they bring new machines to market.

Figure 1: Shotcrete project management dashboard

Data-Logging Inside the Tunnel

Mobbot solutions focus on sprayed concrete operations. Depending on the type of machine, different parameters are monitored and collected using Mobbot’s data-logging devices. These devices can read the machine data either via CAN bus or, in some cases, via an RS-232 interface and even via USB. The data logging box comes in two pre-configurations. One configuration uploads data using an existing Wi-Fi network inside a tunnel. The second connects to the GSM network, and based on the GSM network coverage inside a tunnel, the upload of data can vary.

Depending on the project and user requirements, additional sensors might be needed to monitor other parameters of the shotcrete process. Supplementary sensors can be deployed on the machine and integrated into the system as well. The combination of the machine data with additional devices can provide more insight into the diagnostics and an overview of what happened when problems occur with the machine and the shotcrete process.

Knowing when a malfunction occurred and how parameters behaved at that time is only part of the information. To determine the root of the problem, it is also beneficial to know at which location in the tunnel the problem emerged. To tackle the common challenge of machine position in the tunnel, Mobbot has developed a tablet device to track this information. It is used by operators as a reporting tool to mark in which section of the tunnel they’re working, and track any potential issues faced during the session.

Access to Cloud Data via User-Optimized Dashboards

All collected data ends up stored and processed in a cloud solution and can be displayed on the customer-tailored user dashboard (Fig. 1). In this way, data can be accumulated and anonymized, before finally being processed. The user dashboard can be accessed from anywhere and from multiple types of devices like PCs, laptops, tablets, or phones. This is a new way of collecting data, as up until now, mostly site managers would collect data from different machines manually and then aggregate them if they have sufficient time and tools.

Cost Saving Potentials and Problem Analyses

In the case of project management, the user dashboard provides a short overview of the spraying statistics, to keep track of the events while tunnel construction progresses. For example, to track concrete consumption or accelerator consumption. Just by saving 1% of the accelerator amount, the costs can be reduced by around 50 000 euros per one kilometer of tunnel. Tunnel progress can also be implemented in the dashboard overview (Fig. 2).

Figure 2: Tunnel progression dashboard

It is important to keep track of spraying parameters for diagnostics and understanding the issues with spraying. For example, from the concrete pressure during the spraying, presented in Figure 3, it could be observed that the pressure builds up to 230 bar. The machine itself is programmed to stop all operations if this value of pressure occurs because this value is unusual for this type of machine and signals that there is an issue. In the same figure, the operator's input is monitored as the percentage of the pump. By increasing the pump settings to pump more concrete, the operator increases the concrete flow, but the concrete pressure rises slowly, ending up finally at 230 bar when the machine almost stops. Moreover, as the pressure builds up slowly but steadily, this indicates that the issue more likely comes from the machine itself, than the bad batch of concrete.

In another example from Figure 4, we could see spikes in the concrete flow, for the period around 13:00 on the 18th of August. This could indicate that an operator tried to start spraying, but there was no concrete flow. If it is compared to the accelerator data for the same time window (Fig. 5), one could observe that the accelerator flow at the same time was sufficient and even more than usual, probably due to the variable pressure at the output, caused by the concrete pressure drop. This could indicate that the issue comes from the concrete pump itself and hence narrows the problem for the maintenance team.

Figure 3: Concrete pressure progression

In another example from figure 6, we can observe that there is a concrete flow for a short amount of time for the period around 13:00 on the 18th of August. This could indicate that an operator tried to start spraying, and the concrete started moving in the pipe but suddenly stopped. When compared to the accelerator data for the same time window, presented in figure 7, one could observe that the accelerator flow at the same time was sufficient and even more than usual. This could suggest that there was a variable pressure at the output, caused by the concrete pressure drop. Finally, this could indicate that the issue comes from the concrete pump itself and hence narrows the problem for the maintenance team

The collection of data from the machine could help manufacturers to react quickly to customer needs. With remote access to the machine errors and their history, manufacturers can decrease the reaction time needed to fix the issues. With Exiron, the machine maintenance and parts supplier for Meyco machines, Mobbot developed a maintenance fleet management system, where the machine errors are monitored online. The example in Figure 6 shows the count of the same error over the selected period.

The historical data of the machine errors provides more insights into how an issue started and could indicate where the root cause is. By monitoring all machines in their scope of maintenance, the initial inspection could be done online, without even going to the field and hence prevent the downtime spent in fixing.

Data Collection for Machine Learning

Data collection could be a powerful tool to allow new technologies to penetrate the industry. In recent years, there has been an increase in deployment of machine learning techniques in different industries. One of them could possibly be sprayed concrete. Part of Mobbot’s R&D team is focused on bringing this technology to customers. Two main areas: detection of the root cause of the issue and machine failure forecasting.

Figure 4: Missing concrete flow

Figure 5: Accelerator flow in case of missing concrete flow

Operators and machine maintenance teams struggle to identify fast whether the root of the problem comes from the operator, machine, or the concrete itself. Data collection can be a backbone for solving this problem. To do that, it should be performed over a longer period, and it also requires input from the operators as these issues arise. The first step is to determine from data if the spraying was regular or if there were some issues.

Figure 6: Error monitoring and collection dashboard

Moreover, even though the data is collected, it could be time-consuming to always go through the whole data, and search for issues when they happened. By collecting and analyzing the spraying parameters, it is possible to develop machine learning models that could automatically find spraying patterns that differentiate heavily from the standard spraying that has been done regularly by an operator. This means that the contractors could use this tool to preselect suspicious parameters during the shotcrete session and focus on understanding why they happened, rather than going through all data every day. On the other hand, by developing this tool, the first step of determining whether there was an issue or not, is performed.

Figure 7 presents part of the shotcrete session. Based on concrete flow, concrete pressure, activator flow, activator pressure and the percentage of the concrete pump that is set by the operator (in the figure called concrete_amount_set_percent), it is possible to identify if the part of the shotcrete session fits the regular spraying pattern. The Green background section represents the part of the session that the Machine learning model marked as an anomaly in spraying. Further visual inspection reveals that the part marked as an anomaly had overpressure and unstable concrete pressure and concrete flow, which should suggest that there was an issue with spraying.

Figure 7: Detection of the abnormal spraying pattern (marked with a green background)

Furthermore, this tool could be used to provide information about the productivity of the team and how much time was spent on a non-effective mode of spraying. This information could be useful for optimizing workflows and the operator’s routines.

The detected anomalies new data are coming after each spraying session. After analyzing the productivity, it could be observed that in some cases in a month, the operator spends around 30% of the spraying hours in non-efficient spraying mode. This only indicates the issue when the machine was running but when this type of issue happens it is expensive for the contractors because multiple stakeholders such as material suppliers, machine suppliers and supervisors are involved.

Clogging Prevention

Another example of the machine learning use case could be clogging prevention. When the concrete pipe is clogged, operators lose time to unclog it and the process itself could be painful and unsafe. By forecasting clogging before it happens, it gives an operator a chance to prevent it, which could decrease downtime and improve working conditions.

Spraying parameters with machine error codes could tell if there was a concrete clog in the pipes of the machine. This relation could be leveraged to create a useful dataset for forecasting the event before it happens. In this case, the event that is forecasted is the clogging of the concrete pipe.

Figure 8 presents monitoring of the most important spraying parameters. The forecasting algorithm works in real-time and analyzes each part of the spraying session using sliding windows. The sliding window marked in white in Figure 8 presents the part of the data that is analyzed.

Figure 8: Par of concrete spraying parameters for a timeframe. The white rectangle presents the data window that is analyzed.

Figure 8: Part of concrete spraying parameters for a timeframe.

In Figure 9, based on this sliding window, the forecasting model predicts that clogging will happen in the section marked with a gray background. With further inspection of the gray section, it could be observed that the pressure rises above 230 bar (marked in Fig. 9 as a red vertical line), which indicates clogging, meaning that the forecasting model prediction was true. Furthermore, the difference in time between the analyzed data and the predicted clog is 120 seconds, which should provide an operator with a significant time to react and prevent the pipe from clogging.

This shows that the forecasting of clogging is possible and the Mobbot team will continue further in this direction to bring this technology into use on different tunnel cites. The robustness of these methods is still not good enough to use this technology every day, but currently multiple pilot projects are organized, and future work should improve this issue.

Figure 9: Part of concrete spraying parameters for a timeframe. The white rectangle presents the data window that is analyzed.


As shown in this article, Mobbot solution brings possibilities to multiple sides of the tunneling industry. A connected user dashboard helps to bring transparency to tunneling companies. It enables faster diagnostics and repair of issues. For people building those machines, it could be a huge historical data library, where engineers could understand the behavior of the machines and learn from it. Furthermore, it was shown that this solution can be a starting point for the implementation of the next generation AI technologies. This could help people improve their working conditions and increase safety as well as diagnose issues faster and even prevent some of them.

Probably the last beneficiary of the Mobbot solution is our planet. The consequences are often over-dimensioning and an associated use of more material, which has a negative impact of the CO2 balance of the construction project. They propose further optimization of the mix design and its impact on the overall pumpability, sprayability and performance need to be tracked with data collection, monitoring and analyses.

Learn more about our IoT solutions.

bottom of page