Pile Integrity Testing: How PET AIM Uses AI to Support Pile Length Assessment
Updated: 6 days ago
Pile integrity testing is an important part of foundation quality control, helping engineers evaluate pile continuity, identify potential anomalies, and assess pile length. However, collecting test data is only one part of the process. Interpreting the results can be challenging, particularly when the pile toe is not clearly visible or a project involves hundreds of piles.
When the waveform is ambiguous, engineers may need additional time to review the signal, compare it with construction records, and decide whether the estimated pile length is credible. On large projects, this process can become a significant workload.
PET AIM (AI Measurement) from Piletest addresses this challenge by providing an AI-generated pile-length assessment based on a machine-learning model trained on a large database of pile integrity test results.
By offering an additional reference point during analysis, PET AIM can help engineers review results more efficiently while retaining professional judgment as the basis for final decisions.
Why Is Pile Length Assessment Challenging?
Low-strain pile integrity testing, commonly called PIT, uses stress-wave reflections to evaluate pile continuity and estimate pile length. A light hammer impact generates a wave that travels down the pile and reflects from changes in its properties and from the pile toe.
The recorded waveform, or reflectogram, helps the engineer interpret these reflections.
When the toe reflection is distinct, assessing pile length may be relatively straightforward. However, real-world results do not always look ideal.
Several factors can make interpretation more difficult:
Ambiguous toe reflections: The pile toe may not produce the clearest feature in the waveform. Sometimes there is a second reflection that looks like the toe reflection.
Signal interference: Reflections from changes in pile cross-section or other features can complicate interpretation.
Uncertain expected length: Construction records or planned dimensions may not fully establish the actual as-built length.
Large datasets: A project involving many piles can generate a substantial number of waveforms that require individual review.
Differences in interpretation: Experienced analysts may sometimes disagree about the most plausible location of the toe reflection.
These challenges can slow reporting and make consistent analysis more demanding. An additional assessment method can be valuable when used to support, rather than replace, the engineer's interpretation.
What Is PET AIM (AI Measurement)?
PET AIM is an AI-assisted pile length assessment feature associated with Piletest's PET pile integrity testing system.
Experienced data engineers developed its underlying machine-learning model using historical pile integrity test results analyzed by experienced geotechnical engineers. The model learns patterns in the recorded signals and uses them to predict the likely pile length for a new test result.
Instead of relying exclusively on a manually selected reflection, the AI provides its own length estimate for the engineer to consider alongside the waveform and other available evidence.
This is particularly useful when the toe reflection is not immediately obvious or when an engineer wants an additional reference point before completing the analysis.
To learn more about the testing system, visit PET pile integrity testing equipment.
How Does the AI Model Work?
Machine learning identifies patterns in existing data and uses those patterns to generate a prediction (inference) for the new input.
For pile length assessment, Piletest's published research describes a model trained on a database of approximately 590,000 test results. The research used a subset of the data for training and reserved the remaining results for validation.
The model was developed to predict pile length from test signal information and the planned pile length supplied as an input. The researchers also evaluated its predictions against assessments made by experienced human analysts.
This distinction is important: the AI does not physically measure the pile or independently confirm its as-built dimensions. It generates a prediction based on patterns learned from previously analyzed test results.
The resulting estimate can help the engineer decide where to focus attention and whether the manually assessed pile length deserves another look.
What Did the Published Research Find?
Piletest's published research reported promising results when comparing the model's predictions with expert assessments.
On the validation dataset, the researchers reported a correlation coefficient (R²) of 0.9757 and a mean absolute percentage error of 1.6% relative to the expert-assessed (Test Engineer) pile lengths.
The study also reported that approximately 90% of the model's predictions were within 2% of the expert assessments, while approximately 94% were within 5%. (i.e., for a 20m pile, the PET AIM was less than 1m away from the Expert assessment in 94% of the cases)
These figures describe the model's performance against the expert assessments used in the research. They should not be interpreted as a guarantee that every prediction represents the actual as-built pile length or that the model will perform identically on every project.
The findings nevertheless demonstrate the potential value of AI as an additional assessment tool for engineers working with large volumes of pile integrity test data.
Also note that the Pile Echo Method that ASTM D5882 is based on does not intend to define pile length, yet it has been found statistically to be quite accurate at that. (Read all about the research work behind AIM here)
How Can PET AIM Help When You Have Many Piles to Analyze?
Consider a project where a testing team has collected results from hundreds of piles. Even if the field testing itself is efficient, reviewing every waveform and identifying uncertain toe reflections can still require considerable attention.
PET AIM can help streamline this workflow by providing an AI-generated pile length estimate for consideration during analysis.
1. Add an Independent Reference Point
When the manually assessed pile length is uncertain, the AI prediction provides another estimate to compare with the visible waveform, planned dimensions, and available construction records.
If the two assessments agree and the signal supports the interpretation, the engineer has additional evidence to consider. If they differ substantially, the discrepancy can flag the pile for closer review.
2. Prioritize Uncertain Results
AI-assisted estimates can help identify results that deserve further attention. A difference between the AI prediction and the engineer's assessment does not automatically mean either one is wrong. It indicates that the result may benefit from closer examination.
This can help teams allocate their review time more deliberately instead of treating every pile as equally difficult. (i.e., using additional filtering to remove noise that masks the toe)
3. Support More Consistent Analysis
A machine-learning model applies patterns learned during training to new results. Its predictions can provide a consistent additional reference across a large dataset, helping analysts compare their interpretations against the same model-based approach.
This does not eliminate the need for qualified review, but it can support a more structured analysis process.
4. Make Large Projects Easier to Manage
On projects involving many piles, even small improvements in the interpretation workflow can be valuable. AI-assisted assessment offers a way to supplement manual review and direct attention toward results that require additional investigation.
The practical benefit depends on the quality of the recorded signals, the characteristics of the piles, and how the team incorporates the AI estimate into its existing procedures.
Does PET AIM Replace the Engineer?
No. Treat AI-assisted pile length assessment as a decision-support tool, not a substitute for engineering expertise.
Because the ASTM D5882 test method for pile integrity testing claims to measure pile length, the length figure in the analysis is only an estimate.
A predicted length is not automatically the actual pile length. The estimate must be considered alongside the waveform, assumed stress-wave velocity, planned pile dimensions, construction records, and any other relevant project information.
The original research also acknowledges that AI predictions can be incorrect or inconclusive. Some signals require a more experienced interpretation or further investigation.
When the AI estimate and the engineer's assessment differ, the appropriate response is to investigate the discrepancy rather than automatically accept one result over the other.
For additional help reviewing difficult test results, Piletest provides Second Opinion Support (SOS).
When Should You Consider AI-Assisted Pile Length Assessment?
PET AIM may be particularly useful when:
The toe reflection is difficult to identify confidently.
The engineer wants an additional reference point for the pile length assessment.
A project contains a large number of pile integrity test results.
The testing team wants to identify discrepancies for further review.
Consistency and efficiency are priorities in a high-volume analysis workflow.
It is important to remember that AI cannot recover reliable information from every ambiguous waveform. Some piles will remain inconclusive (The AIM function will issue a failed-to-assess warning), and those results should be reported and investigated appropriately.
Frequently Asked Questions
1. What is PET AIM in pile integrity testing?
PET AIM, or AI Measurement, is an AI-assisted feature that provides a predicted pile length based on a machine-learning model trained on previously analyzed pile integrity test results.
2. How much data was used to train the PET AI model?
Piletest's published research describes a database of approximately 590,000 test results, with a portion used for training and the remainder reserved for validation. The research reported promising agreement between model predictions and expert assessments.
3. Can PET AIM determine the exact as-built pile length?
No. PET AIM generates a model-based estimate rather than a direct physical measurement. This is a known limitation of the PIT method defined in ASTM D5882. Engineers must consider the prediction alongside the waveform, assumed wave velocity, construction information, and other relevant evidence.
4. Is AI-assisted analysis useful when testing hundreds of piles?
It can be. AI-generated length estimates provide an additional reference point that may help engineers review large datasets more efficiently and prioritize uncertain results for closer examination. The final interpretation remains an engineering responsibility.
Conclusion: Combine AI Assistance with Engineering Judgment
Assessing pile length is not always straightforward, especially when toe reflections are unclear or a project generates hundreds of waveforms to review. PET AIM offers an additional way to approach this challenge by using machine learning to generate a pile length estimate based on patterns learned from previously analyzed test results.
For engineers, the value lies in having another reference point, identifying discrepancies that deserve attention, and supporting a more efficient review process.
AI can help make pile integrity test analysis more efficient, but sound engineering judgment remains essential. By combining AI-assisted estimates with careful waveform interpretation and relevant construction information, testing teams can make more informed decisions about pile length and identify results that require further investigation.
Explore PET pile integrity testing to learn more about Piletest's testing and analysis solutions.
Comments