Low Error Research Articles

A Continuous and Noninvasive Method to Estimate Pao2/Fio2 Ratio.

To validate a method for continuously estimating the Pao2/Fio2 (PF) ratio in all critically ill pediatric patients using only standard continuous data monitoring. Retrospective study on a high temporal resolution database. PICU in Montreal, QC, Canada. We included any patients admitted from May 2015 to May 2023 who had an arterial blood gas (ABG) with concurrent continuous pulsed oximetry saturation (Spo2) values. We used our previously validated mathematical model to determine the magnitude of hypoxemia by computing the estimated ePao2/Fio2 (ePF) ratio and comparing it to the Spo2/Fio2 (SF), using PF ratio as the reference standard. None. We analyzed a total of 20,828 ABGs. When Spo2 was below or equal to 97%, the ePF ratio showed a significantly better hypoxemia classification (none, light/moderate, or severe) than the SF ratio (0.80 vs. 0.72; p < 0.001), a lower fixed bias (16.26 vs. -35.24; p < 0.001), a lower mean absolute error (37.92 vs. 63.93; p < 0.001) and a lower proportional bias (slope of 1.01 vs. 0.81; p < 0.001). ePF ratio has also a better limits of agreement difference from Bland-Altman plot (248.10 vs. 292.45; p < 0.001) and coefficient of determination (0.68 vs. 0.59; p < 0.001). When Spo2 was above 97%, the ePF ratio had better classification with Kappa (0.53 vs. 0.43; p < 0.001) and lower fixed bias (-0.63 vs. 65.68; p < 0.001). The PF ratio based on ePF allows for a continuous estimation of hypoxemia severity with a better performance than the SF ratio.

Terahertz Non-Destructive Testing of Porosity in Multi-Layer Thermal Barrier Coatings Based on Small-Sample Data

Accurately characterizing the internal porosity rate of thermal barrier coatings (TBCs) was essential for prolonging their service life. This work concentrated on atmospheric plasma spray (APS)-prepared TBCs and proposed the utilization of terahertz non-destructive detection technology to evaluate their internal porosity rate. The internal porosity rates were ascertained through a metallographic analysis and scanning electron microscopy (SEM), followed by the reconstruction of the TBC model using a four-parameter method. Terahertz time-domain simulation data corresponding to various porosity rates were generated employing the time-domain finite difference method. In simulating actual test signals, white noise with a signal-to-noise ratio of 10 dB was introduced, and various wavelet transforms were utilized for denoising purposes. The effectiveness of different signal processing techniques in mitigating noise was compared to extract key features associated with porosity. To address dimensionality challenges and further enhance model performance, kernel principal component analysis (kPCA) was employed for data processing. To tackle issues related to limited sample sizes, this work proposed to use the Siamese neural network (SNN) and generative adversarial network (GAN) algorithms to solve this challenge in order to improve the generalization ability and detection accuracy of the model. The efficacy of the constructed model was assessed using multiple evaluation metrics; the results indicate that the novel hybrid WT-kPCA-GAN model achieves a prediction accuracy exceeding 0.9 while demonstrating lower error rates and superior predictive performance overall. Ultimately, this work presented an innovative, convenient, non-destructive online approach that was safe and highly precise for measuring the porosity rate of TBCs, particularly in scenarios involving small sample sizes facilitating assessments regarding their service life.

Applying survey weights to ordinal regression models for improved inference in outcome-dependent samples with ordinal outcomes.

Researchers often use outcome-dependent sampling to study the exposure-outcome association. The case-control study is a widely used example of outcome-dependent sampling when the outcome is binary. When the outcome is ordinal, standard ordinal regression models generally produce biased coefficients when the sampling fractions depend on the values of the outcome variable. To address this problem, we studied the performance of survey-weighted ordinal regression models with weights inversely proportional to the sampling fractions. Through an extensive simulation study, we compared the performance of four ordinal regression models (SM: stereotype model; AC: adjacent-category logit model; CR: continuation-ratio logit model; and CM: cumulative logit model), with and without sampling weights under outcome-dependent sampling. We observed that when using weights, all four models produced estimates with negligible bias of all regression coefficients. Without weights, only stereotype model and adjacent-category logit model produced estimates with negligible to low bias for all coefficients except for the intercepts in all scenarios. In one scenario, the unweighted continuation-ratio logit model also produced estimates with low bias. The weighted stereotype model and adjacent-category logit model also produced estimates with lower relative root mean square errors compared to the unweighted models in most scenarios. In some of the scenarios with unevenly distributed categories, the weighted continuation-ratio logit model and cumulative logit model produced estimates with lower relative root mean square errors compared to the respective unweighted models. We used a study of knee osteoarthritis as an example.

Cognitive Radar Waveform Selection for Low-Altitude Maneuvering-Target Tracking: A Robust Information-Aided Fusion Method

In this paper, we introduce an innovative interacting multiple-criterion selection (IMCS) idea to design the optimal radar waveform, aimingto reduce tracking error and enhance tracking performance. This method integrates the multiple-hypothesis tracking (MHT) and Rao–Blackwellized particle filter (RBPF) algorithms to tackle maneuvering First-Person-View (FPV) drones in a three-dimensional low-altitude cluttered environment. A complex hybrid model, combining linear and nonlinear states, is constructed to describe the high maneuverability of the target. Based on the interacting multiple model (IMM) framework, our proposed IMCS method employs several waveform selection criteria as models and determines the optimal criterion with the highest probability to select waveform parameters. The simulation results indicate that the MHT–RBPF algorithm, using the IMCS method for adaptive parameter selection, exhibits high accuracy and robustness in tracking a low-altitude maneuvering target, resulting in lower root mean square error (RMSE) compared with fixed- or single-waveform selection mechanisms.

A Novel Interpolation Method for Soil Parameters Combining RBF Neural Network and IDW in the Pearl River Delta

Soil fertility is a critical factor in agricultural production, directly impacting crop growth, yield, and quality. To achieve precise agricultural management, accurate spatial interpolation of soil parameters is essential. This study developed a new interpolation prediction framework that combines Radial Basis Function (RBF) neural networks with Inverse Distance Weighting (IDW), termed the IDW-RBFNN. This framework initially uses the IDW method to apply preliminary weights based on distance to the data points, which are then used as input for the RBF neural network to form a training dataset. Subsequently, the RBF neural network further trains on these data to refine the interpolation results, achieving more precise spatial data interpolation. We compared the interpolation prediction accuracy of the IDW-RBFNN framework with ordinary Kriging (OK) and RBF methods under three different parameter settings. Ultimately, the IDW-RBFNN demonstrated lower error rates in terms of RMSE and MRE compared to direct RBF interpolation methods when adjusting settings based on different power values, even with a fixed number of data samples. As the sample size decreases, the interpolation accuracy of OK and RBF methods is significantly affected, while the error of IDW-RBFNN remains relatively low. Considering both interpolation accuracy and resource limitations, we recommend using the IDW-RBFNN method (p = 2) with at least 60 samples as the minimum sampling density to ensure high interpolation accuracy under resource constraints. Our method overcomes limitations of existing approaches that use fixed steady-state distance decay parameters, providing an effective tool for soil fertility monitoring in delta regions.

Modelling the properties of aerated concrete on the basis of raw materials and ash-and-slag wastes using machine learning paradigm

The thermal power industry, as a major consumer of hard coal, significantly contributes to harmful emissions, affecting both air quality and soil health during the operation and transportation of ash and slag waste. This study presents the modeling of aerated concrete using local raw materials and ash-and-slag waste in seismic areas through machine learning techniques. A comprehensive literature review and comparative analysis of normative documentation underscore the relevance and feasibility of employing non-autoclaved aerated concrete blocks in such regions. Machine learning methods are particularly effective for disjointed datasets, with neural networks demonstrating superior performance in modeling complex relationships for predicting concrete strength and density. The results reveal that neural networks, especially those with Bayesian Regularisation, consistently outperformed decision trees, achieving higher regression values (Rstrength = 0.9587 and Rdensity = 0.91997) and lower error metrics (MSE, RMSE, RIE, MAE). This indicates their advanced capability to capture intricate non-linear patterns. The study concludes that artificial neural networks are a robust tool for predicting concrete properties, crucial for producing non-autoclaved curing wall blocks suitable for earthquake-resistant construction. Future research should focus on optimizing the balance between density and strength of blocks by enhancing the properties of aerated concrete and utilizing reliable models.

Efficient Environment Map Rendering Based on Decomposition

AbstractThis paper presents an efficient environment map sampling algorithm designed to render high‐quality, low‐noise images with only a few light samples, making it ideal for real‐time applications. We observe that bright pixels in the environment map produce high‐frequency shading effects, such as sharp shadows and shading, while the rest influence the overall tone of the scene. Building on this insight, our approach differs from existing techniques by categorizing the pixels in an environment map into emissive and non‐emissive regions and developing specialized algorithms tailored to the distinct properties of each region. By decomposing the environment lighting, we ensure that light sources are deposited on bright pixels, leading to more accurate shadows and specular highlights. Additionally, this strategy allows us to exploit the smoothness in the low‐frequency component by rendering a smaller image with more lights, thereby enhancing shading accuracy. Extensive experiments demonstrate that our method significantly reduces shadow artefacts and image noise compared to previous techniques, while also achieving lower numerical errors across a range of illumination types, particularly under limited sample conditions.

Noise reduction method for mine wind speed sensor data based on CEEMDAN-wavelet threshold

The mine wind speed sensor is an important intelligent sensing equipment in the mine intelligent ventilation system that can provide accurate and key wind speed parameters for the intelligent ventilation system. The turbulent pulsation characteristics of the airflow in the underground tunnel are a major factor for the inaccurate measurement of mine wind speed. Therefore, according to the random non-stationary characteristics of a turbulent pulsation signal, a denoising method based on adaptive complete ensemble empirical mode decomposition (CEEMDAN) combined with the wavelet threshold is proposed for suppressing the turbulent pulsation noise in the wind speed signal. First, the CEEMDAN algorithm is used for decomposing the wind speed signal into a series of IMF components. Second, the continuous mean square error criterion is used for determining the high-frequency IMF components with more noise. The wavelet threshold denoising method is used for denoising the high-frequency IMF components with more noise. Finally, the denoised IMF components and remaining low-frequency IMF components are reconstructed for obtaining the denoised signal. The results of the denoising analysis of measured turbulent pulsation signals, comparative analysis of denoising of simulated turbulent pulsation signals by different joint denoising methods, and denoising analysis of actual mine wind speed sensor data indicate that the joint denoising method proposed in this study has a higher signal-to-noise ratio and lower root mean square error of the wind speed signal after denoising. Compared with the EMD-wavelet threshold and EEMD-wavelet threshold denoising methods, the denoising method proposed in this study is better and has higher denoising accuracy, which provides a new method for processing actual mine wind speed sensor data.

Forecasting Orange Juice Futures: LSTM, ConvLSTM, and Traditional Models Across Trading Horizons

This study evaluated the forecasting accuracy of various models over 5-day and 10-day trading horizons to predict the prices of orange juice futures (OJ = F). The analysis included traditional models like Autoregressive Integrated Moving Average (ARIMA) and advanced neural network models such as Long Short-Term Memory (LSTM), Recurrent Neural Network (RNN), Backpropagation Neural Network (BPNN), Support Vector Regression (SVR), and Convolutional Long Short-Term Memory (ConvLSTM), incorporating factors like the Commodities Index and the S&amp;P500 Index. We employed loss function metrics and various tests to assess model performance. The results indicated that for the 5-day horizon, the LSTM and ConvLSTM consistently outperformed the other models. LSTM achieved the lowest error rates and demonstrated superior capability in capturing temporal dependencies, especially in single-factor and S&amp;P500 Index predictions. ConvLSTM also performed strongly, effectively modeling spatial and temporal data patterns. In the 10-day horizon, similar trends were observed. LSTM and ConvLSTM models had significantly lower errors and better alignment with actual values. The BPNN model performed well when all factors were included, and the SVR model maintained consistent accuracy, particularly for single-factor predictions. The Diebold–Mariano (DM) test indicated significant differences in forecasting accuracy, favoring advanced neural network models. In addition, incorporating multiple influencing factors further improved predictive performance, enhancing investment outcomes and reducing risk.

Procedure for reducing cross-resonance gate errors using pulse-level control

Abstract Current implementations of superconducting qubits are often limited by the low fidelities of multiqubit gates. We present a reproducible and runtime-efficient pulse-level approach for calibrating an improved cross-resonance gate CR(θ) for arbitrary θ. This CR(θ) gate can be used to produce a wide range of other two-qubit gates via the application of standard single-qubit gates. By performing an interleaved randomised benchmarking sequence, we demonstrate that our approach leads to significantly lower incoherent errors than the circuit-level approach currently used by IBM. Hence, our procedure provides a genuine improvement for applications where noise remains a limiting factor.

Machine Learning for the Genomic Prediction of Growth Traits in a Composite Beef Cattle Population.

The adoption of genomic selection is prevalent across various plant and livestock species, yet existing models for predicting genomic breeding values often remain suboptimal. Machine learning models present a promising avenue to enhance prediction accuracy due to their ability to accommodate both linear and non-linear relationships. In this study, we evaluated four machine learning models-Random Forest, Support Vector Machine, Convolutional Neural Networks, and Multi-Layer Perceptrons-for predicting genomic values related to birth weight (BW), weaning weight (WW), and yearling weight (YW), and compared them with other conventional models-GBLUP (Genomic Best Linear Unbiased Prediction), Bayes A, and Bayes B. The results demonstrated that the GBLUP model achieved the highest prediction accuracy for both BW and YW, whereas the Random Forest model exhibited a superior prediction accuracy for WW. Furthermore, GBLUP outperformed the other models in terms of model fit, as evidenced by the lower mean square error values and regression coefficients of the corrected phenotypes on predicted values. Overall, the GBLUP model delivered a superior prediction accuracy and model fit compared to the machine learning models tested.

Design and Mechanical Validation of Commercially Viable, Personalized Passive Prosthetic Feet

Abstract Current high-performance prosthetic feet work well for many users, but the low resolution of size and stiffness categories may limit walking performance for certain users. A line of prosthetic feet with a high resolution of sizes and stiffnesses, designed through amputee-specific personalization, could provide clinical and economic value. The lower leg trajectory error (LLTE) design framework facilitates the design of high-performance, amputee-specific prosthetic feet; however, previous foot prototypes were not designed to satisfy the economic, mechanical, and aesthetic requirements for commercial adoption. The aims of this work were to understand how a personalized, affordable prosthetic foot can align with the clinical-commercial ecosystem, innovate a viable future product, and inform other prosthesis designers of considerations required to connect innovation to real-world implementation. We evaluated needs by identifying how products, capital, and services flow between stakeholders, and we elucidated design requirements for a personalized prosthetic foot that can be manufactured, distributed, and clinically provided. Based on material properties and manufacturing process capabilities, computer numerically controlled (CNC) machining of Nylon 6/6 satisfies these requirements. We present a novel parametric foot architecture that can be CNC machined, fits within a commercial foot shell, and can be designed for individual users’ body characteristics and activity levels. Prototypes made using the new foot design behaved as anticipated (1–12% error in modeled displacement), satisfied industry-standard strength (ISO 10328) and mechanical performance (AOPA dynamic heel/keel) requirements, and elicited positive feedback from both amputees and prosthetists.

Estimation on compressive strength of recycled aggregate self-compacting concrete using interpretable machine learning-based models

PurposeRecycled aggregate self-compacting concrete (RASCC) has the potential for sustainable resource utilization and has been widely applied. Predicting the compressive strength (CS) of RASCC is challenging due to its complex composite nature and nonlinear behavior.Design/methodology/approachThis study comprehensively evaluated commonly used machine learning (ML) techniques, including artificial neural networks (ANN), random trees (RT), bagging and random forests (RF) for predicting the CS of RASCC. The results indicate that RF and ANN models typically have advantages with higher R2 values, lower root mean square error (RMSE), mean square error (MSE) and mean absolute error (MAE) values.FindingsThe combination of ML and Shapley additive explanation (SHAP) interpretable algorithms provides physical rationality, allowing engineers to adjust the proportion based on parameter analysis to predict and design RASCC. The sensitivity analysis of the ML model indicates that ANN’s interpretation ability is weaker than tree-based algorithms (RT, BG and RF). ML regression technology has high accuracy, good interpretability and great potential for predicting the CS of RASCC.Originality/valueML regression technology has high accuracy, good interpretability and great potential for predicting the CS of RASCC.

Inertial measurement unit technology for gait detection: a comprehensive evaluation of gait traits in two Italian horse breeds.

The shift of the horse breeding sector from agricultural to leisure and sports purposes led to a decrease in local breeds' population size due to the loss of their original breeding purposes. Most of the Italian breeds must adapt to modern market demands, and gait traits are suitable phenotypes to help this process. Inertial measurement unit (IMU) technology can be used to objectively assess them. This work aims to investigate on IMU recorded data (i) the influence of environmental factors and biometric measurements, (ii) their repeatability, (iii) the correlation with judge evaluations, and (iv) their predictive value. The Equisense Motion S® was used to collect phenotypes on 135 horses, Bardigiano (101) and Murgese (34) and the data analysis was conducted using R (v.4.1.2). Analysis of variance (ANOVA) was employed to assess the effects of biometric measurements and environmental and animal factors on the traits. Variations in several traits depending on the breed were identified, highlighting different abilities among Bardigiano and Murgese horses. Repeatability of horse performance was assessed on a subset of horses, with regularity and elevation at walk being the traits with the highest repeatability (0.63 and 0.72). The positive correlation between judge evaluations and sensor data indicates judges' ability to evaluate overall gait quality. Three different algorithms were employed to predict the judges score from the IMU measurements: Support Vector Machine (SVM), Gradient Boosting Machine (GBM), and K-Nearest Neighbors (KNN). A high variability was observed in the accuracy of the SVM model, ranging from 55 to 100% while the other two models showed higher consistency, with accuracy ranging from 74 to 100% for the GBM and from 64 to 88% for the KNN. Overall, the GBM model exhibits the highest accuracy and the lowest error. In conclusion, integrating IMU technology into horse performance evaluation offers valuable insights, with implications for breeding and training.

Exploiting RSM and ANN modeling methods to optimize phosphate ions removal using LDH/alginate composite beads

In the present work, layered double hydroxide (MgAl) and alginate composite beads (LDH/alginate) were developed and used as low-cost and environmentally friendly adsorbent for phosphate ions removal. The successful incorporation of sodium alginate into the LDH structure was confirmed through SEM analysis. The dried beads exhibited a notably rough surface, which promotes molecular movement and potentially enhances pollutant adsorption. The pH of zero charge point (pHPZC) of the composite was found to be 7.41. This result implies that negative ions have a tendency to draw positive charges on the adsorbent surface via electrostatic interaction forces when the pH is lower than the pHPZC. On the other hand, a surface that has a pH higher than pHPZC mostly has a negative charge. The percentage removal and adsorption capacity were investigated as a function of contact time.Experiments were carried out by simultaneously varying three factors, in order to evaluate and compare the predictive capabilities of the response surface methodology (RSM) and the artificial neural network approach (ANN) for the adsorption process. Both methods demonstrated a strong ability to accurately predict the adsorption process. However, the response surface methodology exhibited a lower prediction error compared to the artificial neural network approach.The central composite design within response surface methodology (CCD-RSM) was employed to optimize the experimental conditions for the adsorption process. The model, which considers three factors: adsorbent dose (A), initial concentration (B), and contact time (C), proved to be significant. Among these, the quadratic term (B2) had the most substantial impact on the phosphate ion adsorption rate. The analysis yielded an R2 value of 0.94, indicating an excellent fit to the data. The findings suggest that increasing contact time and reducing the initial concentration improve phosphate ion removal efficiency, while the adsorbent dose has little to no effect. The Artificial Neural Network (ANN) model effectively predicted the adsorptive remediation of phosphate ions onto LDH/alginate composite beads, achieving a high coefficient of determination (R2 = 0.984) between the model outputs and the experimental data. The study highlights the significant potential of LDH/alginate composite beads as a natural adsorbent for the removal of phosphate ions from aqueous solutions. These results underscore the environmental friendliness and efficiency of the adsorbent used for phosphate adsorption.

Hybrid learning-based visual path following for an industrial robot

Abstract This study proposes a novel hybrid learning approach for developing a visual path-following algorithm for industrial robots. The process involves three steps: data collection from a simulation environment, network training, and testing on a real robot. The actor network is trained using supervised learning for 500 epochs. A semitrained network is then obtained at the $250^{th}$ epoch. This network is further trained for another 250 epochs using reinforcement learning methods within the simulation environment. Networks trained with supervised learning (500 epochs) and the proposed hybrid learning method (250 epochs each of supervised and reinforcement learning) are compared. The hybrid learning approach achieves a significantly lower average error (30.9 mm) compared with supervised learning (39.3 mm) on real-world images. Additionally, the hybrid approach exhibits faster processing times (31.7 s) compared with supervised learning (35.0 s). The proposed method is implemented on a KUKA Agilus KR6 R900 six-axis robot, demonstrating its effectiveness. Furthermore, the hybrid approach reduces the total power consumption of the robot’s motors compared with the supervised learning method. These results suggest that the hybrid learning approach offers a more effective and efficient solution for visual path following in industrial robots compared with traditional supervised learning.

A dual-stream temporal convolutional network for remaining useful life prediction of rolling bearings

Abstract Remaining useful life (RUL) prediction plays an indispensable role in the reliable operation and improved maintenance of rolling bearings. Currently, data-driven methods based on deep learning have made significant progress in RUL prediction. However, most of such methods only consider the correlation between channels, ignoring the importance of different time steps for RUL prediction. In addition, it is still challenging to effectively fuse the degradation features of rolling bearings to improve the model RUL prediction performance. To address the above issues, this paper proposes a novel data-driven RUL prediction method named dual-stream temporal convolution network (DSTCN). First, a hybrid attention temporal convolution block (HATCB) is designed to capture the correlation of degraded features on the channel dimension and temporal dimension. Second, a one-dimensional attention fusion module is designed. This module is capable of weight recalibration and assignment to adaptively fuse different degraded features. Afterward, the Hilbert Marginal spectrum is obtained using the Hilbert Huang Transform and used as the input to one stream. Meanwhile, vibration signals are used as the input of the other stream, thus building a dual-stream temporal convolution network to realize RUL prediction. The effectiveness of the proposed method is validated with two life-cycle datasets, and the results show that the method has lower prediction error than other methods for RUL prediction and prognostic analysis.

Multi-fidelity Gaussian process surrogate modeling for regression problems in physics

Abstract One of the main challenges in surrogate modeling is the limited availability of data due to resource constraints associated with computationally expensive simulations. Multi-fidelity methods provide a solution by chaining models in a hierarchy with increasing fidelity, associated with lower error, but increasing cost. In this paper, we compare different multi-fidelity methods employed in constructing Gaussian process surrogates for regression. Non-linear autoregressive methods in the existing literature are primarily confined to two-fidelity models, and we extend these methods to handle more than two levels of fidelity. Additionally, we propose enhancements for an existing method incorporating delay terms by introducing a structured kernel. We demonstrate the performance of these methods across various academic and real-world scenarios. Our findings reveal that multi-fidelity methods generally have a smaller prediction error for the same computational cost as compared to the single-fidelity method, although their effectiveness varies across different scenarios.

A study on generalized balanced split drift stochastic Runge- Kutta methods for stochastic differential equations

To reduce computational complexity, the balanced numerical approximations of the general split drift stochastic Runge-Kutta methods are analyzed. The primary reasons for considering the numerical approximations of these balanced split stochastic Runge-Kutta methods are their improved stability characteristics and lower mean square error compared to other methods. By balancing the drift and diffusion components, the splitting techniques outperform the mean square error over longer time increments. For Ito multi-dimensional stochastic differential equations, we propose a novel family of balanced universal split stochastic Runge-Kutta procedures. The Kronecker product concept is utilized to derive the mean-square stability conditions. We conduct numerical tests to evaluate these methods against an existing weak order 2 split drift method. Ultimately, a specific numerical example validates the theoretical outcomes of the balanced general split stochastic Runge-Kutta procedures.

Power or speed: Which metric is more accurate for modelling endurance running performance on track?

This study aimed to compare the accuracy of the power output, measured by a power meter, with respect to the speed, measured by an inertial measurement unit (IMU) and a global navigation satellite system (GNSS) sport watch to determine the critical power (CP) and speed (CS), work over CP (W') and CS (D'), and long-duration performance (i.e., 60 min). Fifteen highly trained athletes randomly performed seven time trials on a 400m track. The CP/CS and W'/D' were defined through the inverse of time model using the 3, 4, 5, 10, and 20 min trials. The 60 min performance was estimated through the power law model using the 1, 3, and 10 min trials and compared with the actual performance. A lower standard error of the estimate was obtained when using the power meter (CP: 2.7 [2.1-3.3] % and W': 13.8 [10.4-17.3] %) compared to the speed reported by the IMU (CS: 3.4 [2.5-4.3] %) and D': 20.7 [16.6-24.7] %) and GNSS sport watch (CS: 3.4 [2.5-4.3] % and D': 20.6 [16.7-24.7] %). A lower coefficient of variation was also observed for the power meter (4.9 [3.7-6.1] %) Regarding the speed reported by the IMU (10.9 [7.1-14.8] %) and GNSS sport watch (10.9 [7.0-14.7] %) in the 60 min performance estimation, the power meter offered lower errors than the IMU and GNSS sport watch for modelling endurance performance on the track.