The problem of downhole vibrations remains a significant barrier for the drilling industry. An emerging technology based on direct interception of disturbing resonances is now being used to address the issue. Such solutions rely on a new level of understanding with respect to the true bottomhole assembly (BHA) dynamics. Artificial intelligence and machine learning (AI/ML) techniques are now helping deliver the full picture.
While the specific product development in this field belongs in the commercial space, the necessity for more and better data on drilling dynamics is commonplace.
This case study describes a successful solution for data harvesting on a level that is both new and found to meet requirement for the application of AI/ML solutions. The study concludes with examples from analysis that reveal the potential from a true image of everything going on in a drillstring.
Reinventing Drilling Dynamics
Traditional analysis of drilling dynamics frequently references data from downhole accelerometers. However, these systems typically capture a mix of effects rather than the underlying coupling of forces that cause resonances and, consequently, poor performance and downtime from failures.
The fundamentals of these resonances reside in the transfer of torsional and axial forces from surface to the bit. On most rigs, the surface input and reactive loads are recorded at sparse intervals (typically every 10 seconds or longer). Capturing this critical information at a high-fidelity level in this environment presents two critical challenges:
- Signal Integrity: Eliminating data corruption and packet loss from intermittent transfer when sampling frequencies exceeding 1,000 Hz using radio-frequency transmission from a rotating shaft in a rough environment.
- Data Pipeline: While modern sensor assemblies can export structural acoustics and stress waves at extreme sample rates, the bottleneck is the data pipeline and the requirement for a continuous stream of megabytes of time-stamped data every second into a usable database.
With the ambition to capture all ongoing forces in the drilling process with nothing missed, Norwegian specialist company Tomax targeted an active, modern 150-ton automated drilling rig. The rig was built in late 2025 and equipped with a unique strain and torsion measurement unit, or sub, integrated as an extension to the derrick drilling machine.
The sub unit came from an industry leader within the area of control and diagnostics of rotating machinery. The unit was accurately calibrated and capable of operating at very fast sampling rates. Tomax’s specialist created a pipeline from this unit using timestamps applied at critical network nodes and intermediate timestamps calculated via a mirrored pipeline configuration.
This method eliminated data-buffering bottlenecks and multithreading errors. As a result, continuous hours of high-fidelity data, in practical terms the full drilling process, was captured in a columnar database using this pipeline.
The actual drilling and data harvesting was conducted in a geothermal appraisal well through micaceous and granitic gneiss. The string was made up of 5-in. pipe, stabilizer, 6¾-in. steerable positive displacement motor, and an 8½-in. insert bit.
Alternating BHA configurations were run systematically, both with and without a tool configured to intercept vibrations in the 2–15 Hz range. The surface drilling parameters were kept constant. The recording started shallow, allowing the capture of stress and acoustic waves representing the near-bit situation.
Measuring at increasing depths provided data on attenuation and distortion with distance. The 4,000-Hz pipeline streamed data directly into the database, off and on bottom. This also included periods with strong vibrations.
Virtually error free data kept stacking up, and a clear image materialized. Data was then handed over to the ML and analysis phase, including the training of a model to recreate key dynamic characteristics of the system and, hence, verify the data validity.
First Analysis
The current date sources in drilling lack the data that is required for making acceptable analysis of torsional fatigue. This has led to a variety of interpretations of what causes the problem. A similar issue involves the cause of deterioration of drill-bit cutting structures.
The American Society for Testing and Materials (ASTM) dictates the use of the rainflow-cycle counting method as the industry-standard algorithm for breaking down irregular, variable-amplitude stress histories into individual fatigue cycles.
With this process readily available via AI agents and a high-fidelity drilling database at hand, analysis according to the ASTM E1049 rainflow fatigue model could be completed. This method has stringent requirements for data quality and, therefore, doubled as a quality assurance check on the database.
First Results
The methodology presented in this case study is highly comparable to the US Federal Aviation Administration’s Flight Strain Survey (FSS) procedure, the process used to ensure the complete metal structure of an airplane is reliable. Thanks to radical changes in the affordability of data pipelines, high-fidelity databases, powerful processors, and AI agents, this same FSS methodology is now viable for underground drilling.
The plots shown in Fig. 1 use the rainflow fatigue model and compares 4 hours of drilling with three BHAs. The analysis was equal to 115 million calibrated datapoints per trace.
The objective was to study the effect on damage with a stiff system compared with systems with a 2–15 Hz interceptor. The results would reveal what range or frequency caused the actual fatigue, both in torsional direction and axially on the cutting structure.
The graphs in Fig. 1 show:
- BHA 6: 2–15 Hz interceptor in micaceous gneiss (red line)
- BHA 7: Reference BHA (blue line) in same micaceous gneiss and possibly granitic gneiss
- BHA 8: 2–15 Hz interceptor (green line) in granitic gneiss
The ease of processing and producing results from such relatively advanced algorithms offers high expectations for AI/ML techniques in the field of drilling safety and efficiency.
Fig. 2 illustrates how torsional vibrations are recorded for the range up to 1,000 Hz with the relative magnitude expressed in dB.
Nils Reimers, SPE, is director of research development at Tomax in Stavanger. He has more than 35 years of experience in drilling, well technology development, and oilfield operations. Reimers joined Tomax in 2005 and leads research and product development efforts focused on drilling dynamics, vibration mitigation, and performance optimization. Before joining Tomax, he held technical and management positions with Equinor, MHWirth, Aker Solutions, and Altus Intervention. He holds a BS degree in petroleum technology from the University of Stavanger and is a member of SPE.
Ole Andre Høiland specializes in signal optimization for Jotron’s subsidiary AD Navigation in Tønsberg, Norway, where he has been part of the core engineering group since 2018. Holding a background from the University of Agder, his work centers on reliability and performance for portable pilot systems. Alongside his primary role in navigation technology, Høiland has contributed as an expert adviser in the development of the drillstring dynamics data pipeline used for research and development purposes by Tomax.