AI Forecasts Solar Pile Uplift Resistance

AI predicts wind‑uplift resistance of solar PV piles

The new Iranian study shows an artificial neural network can estimate the uplift capacity of driven steel piles for utility‑scale solar farms with a mean absolute percentage error of about 7.6 %. By feeding the model four easy‑to‑measure inputs – pile penetration rate, embedment length, soil friction angle and pile lateral surface area – engineers can predict whether a pile will resist wind‑induced uplift without relying on overly conservative formulas.

The research team tested 130 ST52 steel piles (79 × IPE 140, 31 × IPE 160, 20 × IPE 180) at a solar plant in Rafsanjan, central Iran. Each pile was driven with a hydraulic rig at nine to ten blows per second, and the penetration rate was captured from video, reference marks and installation time. Pull‑out tests increased the force in 500 kgf steps until 20 mm of displacement was reached, defining the uplift capacity. The ANN, trained with 10‑fold cross‑validation and regularised by dropout and early stopping, achieved an R² of 0.78, an RMSE of 0.787 t, and a MAPE of 7.63 %.

Why uplift matters for solar farms

Solar PV structures have a large exposed area but relatively little dead weight, so wind can generate significant uplift and overturning forces. In weak, loose or saturated soils – common at many desert‑side solar sites – the shaft resistance of driven piles may be insufficient, leading to costly redesigns or failures. Traditional static formulas can be over‑conservative because they ignore the dynamic interaction between pile driving, soil friction and installation speed. The ANN therefore offers a data‑driven shortcut that respects site‑specific conditions while keeping safety margins.

Key predictors uncovered by SHAP analysis

Interpretability via Shapley additive explanations (SHAP) showed that penetration rate (≈ 49.8 %) and soil friction angle (≈ 47.1 %) dominate the model’s predictions, while geometric parameters – embedment length and lateral surface area – play a secondary role. This aligns with field observations that faster driving can compact the surrounding sand, increasing shaft resistance, and that higher friction angles (i.e., more granular, dry soils) boost uplift resistance.

How the model compares to other studies

Other recent pile‑uplift research, such as a full‑scale field study in India, reported an R² of 0.94 for steel piles but used a larger, multi‑site dataset and different soil types). The Iranian ANN’s slightly lower R² reflects its single‑site focus on a sandy environment, yet its 7.6 % MAPE is competitive for early‑stage design tools where rapid decisions are needed.

Practical takeaways for engineers

1. Collect four simple inputs – penetration rate, embedment length, soil friction angle, and pile lateral area – during pile driving.
2. Run the ANN (available on request) to obtain a predicted uplift capacity at 20 mm displacement.
3. Validate with a limited number of pull‑out tests to confirm the model’s applicability to local soil conditions.

What it means for Israel

Some Israeli solar farms are installed on arid, sandy soils, which may share characteristics with the site studied in Iran, making the findings potentially relevant. Using the typical Israeli residential tariff of ₪0.48/kWh and a turnkey installation cost of ₪3,150/kWp, a 10 kWp rooftop system would generate roughly 17,000 kWh/year (central‑region yield of 1,700 kWh/kWp). That translates to ₪8,160 of annual revenue, paying back the ₪31,500 investment in about 3.9 years. If a utility‑scale farm adopts the ANN‑based design, it could reduce pile material costs by avoiding over‑design, supporting the nation’s goal of 30 % renewable electricity by 2030.

Forward look

The authors stress that the model must be tested on varied geologies – clay, silt, gravel – before becoming a universal tool. Future work will likely integrate additional parameters (e.g., moisture content, pile hammer energy) and expand the dataset across multiple solar sites in the region. For Israeli developers, early adoption of AI‑driven pile design could accelerate project timelines, lower capital expenditures, and help meet the country’s ambitious renewable targets.

---

Sources: PV Magazine, Springer Nature, EPCLand – Solar Pile Design Data

Sources & further reading

• AI predicts wind-uplift resistance of solar PV piles - pv magazine Global
• Short Piles for a Solar Power Plant in Western Rajasthan
• Amirkabir Journal of Civil Engineering - Author Index
• Effect of Installation Method on the Uplift Capacity of Single Driven Steel Piles: Full-Scale Field Study | Indian Geotechnical Journal | Springer Nature Link
• Essential Geotechnical Pile Design Data for Utility-Scale Solar Structures