BTS | HARDING PRIZE COMPETITION 2023
Table 6: Weekly average projection Projection
Actual
Neural Network Benchmark
124.235 105.4 70
Average Rings Per Week
training phase to ensure the algorithms had more
exposure on variable data. The output of the neural network is shown in Figure 9. The model predicted high spikes, due to its
anticipation of harder excavation from the sandstone and gypsum. This helped Murphy Group to prepare for these uncertainties. Interestingly, it was also predicting lower ring cycle durations than the proposed average. The model also predicted that the sandstone would be difficult to adjust to, noticeable between Ring 5033 and 5190. The behaviour of the machine in sandstone was accurately anticipated as it was reported that the sandstone was significantly harder to excavate yet was predicted with little available data. Figure 10 shows the predictions against the actual
results. Here, the weekly forecast from the neural network
was able to anticipate a decrease in productivity as the project ended yet maintained a weekly average that was often higher than the benchmark weekly production rate from Murphy Group. Although more data would help to further improve the predictive ability of the neural network, the forecast was significantly more accurate than the given benchmark, received on the same date as the neural network on 13 September 2021. Whilst the existing benchmark predicted a finish
in May, the neural network predicted a finish on 14 February 2022, within 14 days after the actual tunnel completion.
CONCLUSION Neural networks demonstrated that they were able to understand the complex nature of tunnelling, and were able to yield accurate predictions, even with limited data. This was shown on two tunnel boring projects, in different ground conditions and with different datasets, demonstrating how generalisable the solution is. This is promising, as patterns could be anticipated in new settings, helping to form a more accurate tender, or to understand productivity as the tunnel is being excavated. Having a more accurate tender helps with budgeting, whilst being able to anticipate risk helps a project manager prepare for a mitigation. Now that the neural network model is built, it can
be applied to future groundwork projects to forecast the TBM performance in new ground conditions. Murphy Group were satisfied with the results
presented from the neural networks studies and is among the first to use machine learning in a tunnelling environment, as well as groundworks in general. Significant innovations in the field of artificial intelligence were made, as the current models have never been used to forecast such an ambitious output. Previously, machine learning researchers would rely on TBM performance data to help with the predictions. The approach proposed showed an innovative method that only used data inputs available at pre-tender.
Above, figure 11: AI studies are advancing possibilities for forecasting tunnelling progress 40 | August 2023
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49
