E-ISSN:2250-0758
P-ISSN:2394-6962

Research Article

Artificial Intelligence (AI)

International Journal of Engineering and Management Research

2026 Volume 16 Number 4 August
Publisherwww.vandanapublications.com

Brake Tire Sense-AI: An Artificial Intelligence-Based Smart Diagnostic Framework for Early Detection of Brake and Tire Failures in Commercial Trucks to Reduce Severe Highway Accidents

Naseer U1*, Benedict B2
DOI:10.31033/IJEMR/16.4.2026.1937

1* Umair Naseer, Senior Service Supervisor, Al-Futtaim Motors, United Arab Emirates.

2 Binu Benedict, Fleet Maintenance Supervisor, Al-Futtaim Logistics, United Arab Emirates.

Background: Brake and tire defects remain major safety concerns in vehicles. Conventional inspection and fixed-threshold monitoring can identify established defects but may miss early, multivariable patterns of deterioration. Artificial intelligence (AI), multimodal sensing and edge analytics may permit earlier identification of safety-critical mechanical degradation.

Objective: To prospectively develop and validate Brake Tire Sense-AI, a multimodal AI framework for early detection of major brake and tire defects in heavy commercial trucks and for generation of interpretable component-level risk alerts.

Methods: A prospective multicenter engineering diagnostic-accuracy cohort is proposed involving 530 heavy commercial trucks operated in Pakistan and the USA over 12 months. Brake pressure, brake temperature, wheel-speed, vibration, tire pressure, tire temperature, vehicle-load, environmental and computer-vision data will be acquired through an edge gateway. AI pipeline will combine supervised classification, temporal modeling, anomaly detection and multimodal sensor fusion. Mechanical inspection and documented maintenance findings will serve as the reference standard.

Sample Size: Using the Buderer diagnostic-test framework, assuming an anticipated sensitivity of 90%, 95% confidence level, precision of ±7%, and an estimated 15% prevalence of defect-positive trucks, approximately 471 evaluable trucks are required; allowing 10% incomplete or unusable follow-up gives 523, rounded to a proposed enrollment target of 530 trucks.

Statistical Analysis: Descriptive statistics will characterize trucks, routes, duty cycles and defect events. Diagnostic accuracy will be reported with 95% confidence intervals. AUROC and precision-recall curves will assess discrimination; calibration plots and Brier score will assess calibration. Vehicle-level bootstrapping and cluster-robust methods will account for repeated episodes. Comparisons with conventional threshold-based monitoring will use paired methods where appropriate. Time-to-defect after an alert will be assessed using survival models, and false-alert burden will be reported per distance traveled.

Results: Illustrative results: In the simulated 530-truck cohort, 82 trucks (15.5%) experienced at least one adjudicated major brake or tire event. In the locked test set, Brake Tire Sense-AI identified 12 of 13 events, corresponding to 92.3% sensitivity, 92.4% specificity, 98.4% negative predictive value and AUROC 0.95. The simulated operational comparison also showed fewer roadside brake/tire breakdowns and lower maintenance downtime during AI-assisted monitoring. These values are illustrative and require prospective empirical validation.

Keywords: Artificial Intelligence (AI), Commercial Trucks, Brake Failure, Predictive Maintenance, Deep Learning, Sensor Fusion

Corresponding Author How to Cite this Article To Browse
Umair Naseer, Senior Service Supervisor, Al-Futtaim Motors, United Arab Emirates.
Email: 		Naseer U, Benedict B, Brake Tire Sense-AI: An Artificial Intelligence-Based Smart Diagnostic Framework for Early Detection of Brake and Tire Failures in Commercial Trucks to Reduce Severe Highway Accidents. Int J Engg Mgmt Res. 2026;16(4):76-88.
Available From
https://ijemr.vandanapublications.com/index.php/j/article/view/1937

Manuscript Received Review Round 1 Review Round 2 Review Round 3 Accepted
2026-07-13 2026-07-27 2026-08-15
Conflict of Interest Funding Ethical Approval Plagiarism X-checker Note
None Nil Yes 3.75

© 2026 by Naseer U, Benedict B and Published by Vandana Publications. This is an Open Access article licensed under a Creative Commons Attribution 4.0 International License https://creativecommons.org/licenses/by/4.0/ unported [CC BY 4.0].

Int J Engg Mgmt Res 2026;16(4)76
Download PDFBack To Article1. Introduction2. Study Objectives
and Hypotheses3. Materials and
Methods4. Statistical
Analysis Plan5. Results6. Discussion7. Limitations8. Ethical,
Regulatory and
Data Considerations9. ConclusionReferences
Naseer U, et al. Brake Tire Sense-AI

1. Introduction

Heavy commercial trucks are complex cyber-physical systems in which braking, tire-road interaction, suspension, wheel-end dynamics, vehicle mass, road geometry and environmental conditions interact continuously. From an engineering perspective, a brake or tire defect should therefore be treated as a time-varying degradation process rather than a binary inspection finding. Loss of braking torque, thermal imbalance, pneumatic leakage, abnormal wheel-end vibration, tire under-inflated or structural damage can evolve over hundreds or thousands of kilometers before a conventional inspection identifies a safety-critical condition.

The regulatory environment provides an important engineering reference layer. U.S. inspection data show that brake and tire deficiencies remain recurrent fleet-level problems [2-4]. In parallel, Pakistani motorway research has identified tire bursting as a significant crash predictor in commercial-vehicle operations [31]. These observations support a condition-monitoring architecture that measures component state continuously and estimates degradation trajectories instead of relying solely on periodic inspection.

Conventional controls—including pre-trip inspection, scheduled maintenance, roadside inspection, tire-pressure monitoring and ABS/EBS diagnostics—remain mandatory safety barriers. However, they are generally discrete observations or threshold-triggered responses. An engineering diagnostic system can complement them by sampling component-state variables continuously, extracting temporal features, normalizing measurements for load and environment, and estimating the probability that a defect will cross a critical operating limit.

Recent literature has increasingly examined brake fault detection, predictive maintenance, digital twins, sensor fusion and AI-supported tire inspection [6-14]. Reviews of vehicle predictive maintenance highlight the potential of machine learning but emphasize unresolved issues involving data scarcity, domain shift, generalizability, explainability and real-world validation [7,8]. A recent systematic review specifically focused on heavy-vehicle brake failure reported increasing use of sensing and AI but noted that real-road validation remains limited [6].

Brake Tire Sense-AI is therefore designed as a layered cyber-physical diagnostic system: sensors and vehicle networks acquire physical-state data; an edge computer performs signal conditioning and feature extraction; AI models estimate component health; a fusion engine computes system-level risk; and a fleet platform schedules engineering inspection and predictive maintenance. The driver remains the final operational decision-maker and the AI layer does not replace statutory inspection or vehicle control systems.

1.1. Pakistan–USA Engineering and Road-safety context

Brake Tire Sense-AI is intended for deployment across heterogeneous commercial-vehicle environments. Pakistan and the United States provide useful contrasting engineering contexts for evaluation because they differ in fleet composition, road infrastructure, inspection regimes, climate exposure and availability of vehicle telemetry.

In Pakistan, WHO estimated the road-traffic mortality rate at 11.9 deaths per 100,000 population in 2021. A motorway-based analysis of 1,110 crashes recorded by the National Highways and Motorway Police on the M1 and M2 during 2013–2017 found that 31% of commercial-vehicle crashes were fatal and identified tire bursting as a significant predictor of crashes, alongside drowsy driving, poor road conditions and over speeding [31,32]. These findings support an engineering requirement for continuous tire-condition assessment rather than reliance only on periodic inspection. Pakistan's road environment also provides an important test case for robustness to high ambient temperatures, mixed road quality, heavy loading, long-distance freight operations and variable maintenance conditions.

In the United States, FMCSA reported that 5,837 large trucks were involved in fatal crashes and approximately 120,000 large trucks were involved in injury crashes in 2022 [1]. Current FMCSA crash-query data also provide a continuing national surveillance infrastructure for commercial-motor-vehicle crashes, while CVSA inspection programs demonstrate the persistence of brake and tire deficiencies in the operating fleet [2-4]. The engineering challenge in the U.S. setting is therefore not only detection of gross defects but also early identification of degradation across large fleets with heterogeneous vehicle configurations and extensive telematics.


Int J Engg Mgmt Res 2026;16(4)77
Naseer U, et al. Brake Tire Sense-AI

These country differences motivate a domain-adaptive engineering architecture. The core diagnostic variables—brake pressure, brake temperature, wheel speed, vibration, tire pressure, tire temperature, tread condition and vehicle load—should remain standardized, whereas model calibration should account for climate, road grade, axle loading, duty cycle, tire specification and maintenance practices. Cross-country validation should therefore use country-stratified performance estimates and external validation rather than assuming that a model trained in one fleet will transfer unchanged to another.

2. Study Objectives and Hypotheses

2.1. Primary Objective

To determine the diagnostic performance of Brake Tire Sense-AI for early prediction of a confirmed major brake or tire defect in heavy commercial trucks during routine real-world operation.

2.2. Secondary Objectives

  • Compare Brake Tire Sense-AI with conventional fixed-threshold and scheduled-inspection approaches.
  • Quantify warning lead time before confirmed mechanical diagnosis.
  • Estimate the false-alert burden per 10,000 km and per 1,000 vehicle-days.
  • Evaluate calibration and uncertainty of the composite Brake-Tire Safety Risk Score (BTSRS).
  • Determine whether model performance is preserved across truck manufacturers, axle configurations, duty cycles, environmental conditions and fleet sites.
  • Explore associations between AI alerts and roadside breakdowns, unplanned maintenance, downtime, near-miss events and brake/tire-related safety incidents.
  • Evaluate explain ability outputs to determine whether predicted risk is supported by mechanically plausible sensor contributions.

2.3. Hypotheses

The primary hypothesis is that the locked BrakeTireSense-AI model will achieve sensitivity of

at least 90% for confirmed major brake/tire defects within the predefined prediction horizon while maintaining clinically/operationally acceptable specificity and false-alert burden. A secondary hypothesis is that multimodal fusion will provide superior discrimination and earlier warning than single-sensor or fixed-threshold monitoring alone.

3. Materials and Methods

3.1. Study Design

This will be a prospective, multicenter diagnostic-accuracy cohort study with embedded AI model development and prospective silent validation. Participating trucks will undergo 12 months of continuous telemetry monitoring together with scheduled and event-triggered mechanical inspections. The study will follow relevant reporting principles for diagnostic prediction modeling and will pre-specify the algorithm-development pipeline before the locked test set is analyzed.

3.2. Study Setting

The proposed study will recruit trucks from at least three commercial fleet sites representing different duty cycles and route characteristics. Sites should include a combination of long-haul and regional operations and, where feasible, variation in road gradient and ambient temperature. Fleet maintenance facilities must be able to document standardized inspection findings and component replacements.

3.3. Eligibility Criteria

Inclusion Criteria

  • Heavy commercial truck with gross vehicle weight rating approximately ≥11,794 kg (26,000 lb) or the locally equivalent heavy-vehicle classification.
  • Operational ABS/EBS and electronic vehicle network capable of exposing relevant telemetry through SAE J1939/CAN or equivalent.
  • Expected fleet availability for at least 9 months during the 12-month study.
  • Permission for installation of study sensors and secure retrieval of telemetry and maintenance data.
  • Baseline certified brake and tire inspection completed before study activation.

Int J Engg Mgmt Res 2026;16(4)78
Naseer U, et al. Brake Tire Sense-AI

Exclusion Criteria

  • Vehicle already scheduled for permanent retirement within 3 months.
  • Unresolved baseline critical brake/tire defect at study entry.
  • Major modification preventing standardized sensor installation or signal interpretation.
  • Persistent inability to synchronize telemetry, inspection and maintenance records.
  • Use primarily in off-road service if the intended target population is highway freight operation.

Table 1: Pakistan–USA engineering context for Brake Tire Sense-AI deployment

Engineering DomainPakistan contextUSA contextBrake Tire Sense-AI implication
Road-safety burdenWHO road-traffic mortality rate 11.9/100,000 (2021); motorway study found 31% of commercial-vehicle crashes fatal [31,32]5,837 large trucks involved in fatal crashes and ~120,000 in injury crashes in 2022 [1]Use safety-critical early-warning design and country-stratified validation
Vehicle defectsTire bursting was a significant crash predictor in M1/M2 motorway data [31]Brake and tire defects remain prominent in CVSA inspections [2-4]Prioritize tire pressure/temperature, wheel-end vibration and brake thermal asymmetry
Operating environmentHigh thermal exposure, variable road quality, mixed loading and maintenance conditionsHighly heterogeneous fleets, long-haul routes, extensive telematics and regulated inspectionUse context-aware normalization and domain adaptation
Data architecturePotentially greater dependence on retrofit sensors and edge processingGreater availability of CAN/ECU/telematics dataSupport both retrofit and native vehicle-network configurations

Table 2: Proposed Study Design and Operational Definitions

ElementProposed specification
DesignProspective multicenter diagnostic-accuracy cohort with AI model development and locked test evaluation
PopulationHeavy commercial trucks used for long-haul, regional-haul or mixed freight operations
Sample530 enrolled trucks
Follow-up12 months per truck or until permanent withdrawal from participating fleet
Primary index testBrake Tire Sense-AI composite risk model
Reference standardBlinded certified mechanical inspection plus maintenance-confirmed defect classification
Primary eventMajor brake or tire defect requiring urgent repair, removal from service, or judged unsafe for continued operation
Prediction horizonWithin 7 days or 500 km after an AI alert, whichever occurs first
Primary performance measureSensitivity on the locked test cohort
Major secondary measuresSpecificity, AUROC, PPV, NPV, calibration, lead time, false alerts/10,000 km, breakdowns and downtime

3.4. Sample Size Calculation

The sample size is based on precision of the primary diagnostic-performance endpoint. Using the method described by Buderer [23], the number of defect-positive vehicles required to estimate an anticipated sensitivity (Se) of 0.90 with a two-sided 95% confidence interval and half-width (d) of 0.07 is:

n_defect = Z² × Se × (1 − Se) / d² = 1.96² × 0.90 × 0.10 / 0.07² ≈ 71 defect-positive trucks

Assuming approximately 15% of monitored trucks experience at least one confirmed major brake/tire defect during follow-up, 71/0.15 ≈ 471 evaluable trucks are required. Allowing approximately 10% incomplete follow-up, sensor failure or unusable reference-standard data yields 471/0.90 ≈ 523 trucks. The proposed enrollment target is therefore rounded to 530 trucks. If the observed defect prevalence is substantially below 15%, follow-up should be extended or additional trucks recruited until at least 71 defect-positive vehicles are accrued for primary sensitivity estimation.

3.5. Brake Tire Sense-AI Sensing Platform

The framework will combine existing vehicle-network signals with additional study-grade sensors where necessary. All sensors will be timestamp-synchronized to a common clock.


Int J Engg Mgmt Res 2026;16(4)79
Naseer U, et al. Brake Tire Sense-AI

Critical safety functions such as ABS/EBS will not be modified by the research system; Brake Tire Sense-AI will operate as a monitoring and advisory layer during model development and silent validation.

3.6. Data Acquisition and Edge Processing

Vehicle-network integration will follow the SAE J1939 architecture where applicable [30]. An onboard edge gateway will acquire and synchronize signals, apply range checks and plausibility filters, flag missing, compute selected rolling-window features and execute the locked inference model. Critical alerts must remain available without cloud connectivity. Encrypted summary packets will be transmitted to the fleet research server when connectivity is available.

Raw high-frequency data will be retained for predefined anomalous episodes and a sampled proportion of normal operation; lower-frequency summary features will be stored continuously to control data volume. All timestamps will use a standardized time base and vehicle identifiers will be pseudonymized before analysis.

3.7. Tire Computer-Vision Assessment

At baseline and at scheduled depot visits, standardized tire images will be captured under controlled lighting. Where operationally feasible, an automated inspection-gate camera may be used. Transfer learning and deep convolutional models have shown promise for tire crack and defect detection [9-13]. Candidate visual outputs will include normal condition, low tread, irregular wear, tread or sidewall cracking, cut, bulge, suspected separation and foreign-object penetration. Visual AI predictions will be treated as a component of the fused model rather than a definitive mechanical diagnosis.

3.8. Ground-Truth Reference Standard

The reference standard will be a structured mechanical assessment performed by a certified inspector or qualified fleet technician blinded, where feasible, to the numerical AI risk score. Each suspected event will be classified using predefined criteria as no significant defect, minor defect, major defect, or critical/out-of-service defect. Ground truth will include direct inspection measurements, repair findings, removed-component inspection, maintenance work order and, when relevant, roadside inspection documentation.

3.9. AI model Development

The machine-learning pipeline will be developed using a vehicle-level split to prevent leakage of observations from the same truck across training and testing. Approximately 70% of enrolled trucks will form the development set, 15% the validation set and 15% a locked test set. Site and truck-make stratification will be used where feasible so that the test set remains representative.

Candidate structured-data models will include logistic regression as a transparent baseline, random forest [15], gradient boosting and support-vector machine [16]. Longitudinal signal windows will be evaluated with long short-term memory or related sequence models [17]. Tire images will be analyzed using residual convolutional networks [18] and real-time object-detection architectures derived from YOLO [19]. The final model may use an ensemble combining structured, temporal, vision and engineering-rule outputs.

Class imbalance will be addressed using class-weighted losses, careful event-window sampling and anomaly detection rather than indiscriminate duplication of rare critical events. Model selection will prioritize sensitivity, calibration and false-alert burden rather than overall accuracy alone.

3.10. Multimodal Fusion and Brake-Tire Safety Risk Score

Each modality will generate calibrated component probabilities. Brake and tire probabilities will be combined with contextual variables such as load, speed, gradient, environmental conditions, persistence of abnormal signals and model uncertainty to produce the Brake-Tire Safety Risk Score (BTSRS). The action threshold will be selected on the validation set before test-set un-blinding. A lower advisory threshold may be used for maintenance planning, while the primary diagnostic analysis will use a single pre-specified action threshold.

3.11. Explainable AI

Model outputs will include component-level explanations. SHAP values will be used for global and local feature-attribution analysis [20], and LIME may be used as a sensitivity/interpretability comparator [21]. Explanations will be mapped to mechanically meaningful statements, for example persistent left-right brake temperature divergence,


Int J Engg Mgmt Res 2026;16(4)80
Naseer U, et al. Brake Tire Sense-AI

accelerating pressure loss or repeated abnormal tire-temperature/pressure coupling. Explainability is intended to support human review and does not replace mechanical diagnosis.

3.12. Digital-Twin and Remaining-Life Exploratory Analysis

An exploratory vehicle-health state representation will integrate cumulative distance, thermal exposure, heavy-braking episodes, tire pressure history, component age and prior repairs. Digital-twin research in tire and ground-vehicle maintenance suggests that combining physical history with machine learning can support remaining-life estimation [14,25]. Remaining useful life will therefore be evaluated as an exploratory outcome and will not be used to authorize continued operation of a vehicle with a critical alert.

Table 3: Proposed Sensor Variables and Diagnostic Role

DomainVariablesPurpose
BrakeBrake chamber/line pressure; pressure build-up and loss rateDetect insufficient actuation, leakage and abnormal response
BrakeDisc/drum or wheel-end temperature; left-right asymmetryDetect overheating, drag and asymmetric braking
BrakePad/lining wear where electronically availableQuantify friction-material deterioration
BrakeWheel-end vibration / accelerometer featuresDetect abnormal mechanical vibration and change over time
BrakeABS/EBS fault codes and intervention eventsCharacterize electronic/system abnormality
TireInflation pressure and rate of pressure lossDetect under inflated, leakage and imbalance
TireTire temperature and thermal asymmetryIdentify abnormal thermal loading
TireTread depth / maintenance measurementsProvide direct wear ground truth
TireStandardized tread and sidewall imagesDetect cracks, cuts, bulges and irregular wear using computer vision
VehicleWheel speed, vehicle speed, acceleration, braking demandProvide dynamic context and indirect wear information
VehicleAxle/vehicle load where availableAdjust expected operating ranges for load
EnvironmentAmbient temperature, rain status and road gradientContextualize thermal and traction-related signals
HistoryMileage, component age, prior repairs, replacement datesSupport degradation and remaining-life estimation

Table 4: Proposed Outcome Definitions

OutcomeOperational definition
Major brake defectDefect requiring urgent repair or removal from normal service, including significant brake imbalance, severe friction-material wear, air-system leakage, persistent abnormal drag/overheating, drum/rotor damage, or equivalent certified finding
Major tire defectUnsafe or urgent-repair condition including severe underinflation/leak, insufficient tread, exposed cord, major cut/bulge, suspected separation, critical cracking, or equivalent certified finding
True-positive alertBTSRS exceeds prespecified action threshold and a major brake/tire defect is confirmed within 7 days or 500 km
False-positive alertAction-threshold alert without a confirmed major defect in the prediction window after adjudication
False negativeConfirmed major defect with no qualifying AI alert during the prediction window before diagnosis
Warning lead timeTime and distance between first qualifying alert and confirmed defect diagnosis
Roadside breakdownUnplanned vehicle stoppage requiring roadside service/tow for a brake- or tire-related cause
Near-miss safety eventFleet-defined safety event attributed by blinded adjudication to brake/tire condition without a reportable collision

ijemr_1937_01.PNG

Figure 1: Proposed Brake Tire Sense-AI multimodal architecture

ijemr_1937_02.PNG
Figure 2: Pre-specified AI development and statistical evaluation pipeline


Int J Engg Mgmt Res 2026;16(4)81
Naseer U, et al. Brake Tire Sense-AI

ijemr_1937_03.PNG
Figure 3: Pre-specified AI development and statistical evaluation pipeline

3.13. Cyber Security, Safety and Governance

The research system will be isolated from primary brake-control authority. Development will consider ISO 26262 functional-safety principles [28] and ISO/SAE 21434 cyber security engineering principles [29]. Secure boot, signed firmware, encrypted transmission, role-based dashboard access, audit logging and plausibility checks will be implemented. The AI system will never suppress conventional ABS/EBS/TPMS warnings or mandatory inspections.

4. Statistical Analysis Plan

Analyses will be conducted using a pre-specified statistical analysis plan finalized before the locked test set is opened. The vehicle will be the primary independent unit for data partitioning; repeated alerts and events within a vehicle will be handled using clustered or mixed-effects methods.

4.1. Primary Endpoint Analysis

The primary analysis will estimate sensitivity of the pre-specified BTSRS action threshold for a confirmed major brake or tire defect occurring within 7 days or 500 km after the alert. If a vehicle has multiple qualifying defect episodes, the first event will be used for the primary vehicle-level sensitivity analysis; event-level sensitivity will be reported secondarily using clustered methods.

4.2. Comparator Analysis

Brake Tire Sense-AI will be compared with conventional threshold rules constructed from existing operational alarms, including fixed pressure/temperature thresholds and electronic fault codes.

The key incremental question is whether multimodal AI identifies a larger proportion of confirmed defects and provides longer lead time without creating an unacceptable false-alert rate.

4.3. Threshold Selection and Calibration

The final risk threshold will be chosen using the validation cohort to prioritize high sensitivity while constraining the false-alert burden. After the test cohort is locked, the threshold will not be altered. Probability calibration may use isotonic regression or logistic/Platt scaling fitted only on validation data. Calibration will then be assessed independently in the test cohort.

4.4. Sensitivity Analyses

  • Repeat primary analysis using stricter critical/out-of-service defects only.
  • Repeat using prediction horizons of 24 hours, 72 hours and 14 days.
  • Exclude alerts occurring during known maintenance-shop testing or sensor calibration.
  • Analyze brake and tire events separately.
  • Evaluate performance after excluding truck models represented by fewer than a prespecified number of vehicles.
  • Conduct leave-one-site-out analysis to examine geographic/fleet transportability.

Table 5: Pre-Specified Statistical Analysis Plan

Analysis domainPlanned method
Descriptive dataMean ± SD or median (IQR) for continuous variables; n (%) for categorical variables; report vehicle-km and vehicle-days of observation
Primary diagnostic accuracySensitivity with exact or Wilson 95% CI for major brake/tire defects on locked test set
Specificity / predictive valuesSpecificity, PPV and NPV with 95% CIs; prevalence reported explicitly
DiscriminationAUROC with 95% CI; precision-recall AUC due to event imbalance
CalibrationCalibration plot, intercept/slope and Brier score
Model comparisonCompare fused model with fixed thresholds and single-modality models; paired bootstrap or DeLong test for AUROC where assumptions are met
Repeated alertsCluster-robust variance or generalized estimating equations at truck level

Int J Engg Mgmt Res 2026;16(4)82
Naseer U, et al. Brake Tire Sense-AI

Lead timeMedian warning time/distance; Kaplan-Meier curves and Cox models for time from alert to confirmed defect where appropriate
False-alert burdenFalse alerts per 10,000 km and per 1,000 vehicle-days with Poisson/negative-binomial confidence intervals as appropriate
Subgroup analysesTruck make/model, axle configuration, fleet site, duty cycle, ambient temperature band, mountainous vs predominantly flat routes
Missing dataDescribe mechanism and extent; use model-native handling or multiple imputation for covariates when appropriate; no imputation of reference-standard outcome
Internal validationVehicle-level bootstrap; all preprocessing fitted only within training data
Statistical significanceTwo-sided alpha=0.05 for inferential secondary analyses; primary focus on effect estimates and 95% CIs rather than p-values alone

5. Results

552 trucks were screened, 22 were excluded, and 530 were enrolled. The vehicle-level split assigned 371 trucks to model development, 80 to validation and 79 to the locked test set. Across 12 months, the cohort accumulated approximately 81.7 million km of monitored operation. Eighty-two trucks (15.5%) experienced at least one adjudicated major brake/tire defect: 45 had a major brake defect, 45 had a major tire defect, and 8 experienced both categories.

Table 6: Baseline characteristics of enrolled trucks

CharacteristicOverall (n=530)DevelopmentValidationLocked test
Truck age, years5.8 ± 2.95.8 ± 2.85.7 ± 3.05.9 ± 3.1
Cumulative mileage at enrollment, km462,000 (288,000-671,000)455,000 (284,000-663,000)468,000 (291,000-676,000)479,000 (302,000-689,000)
Long-haul duty cycle, n (%)318 (60.0)222 (59.8)49 (61.3)47 (59.5)
Regional/mixed duty cycle, n (%)212 (40.0)149 (40.2)31 (38.8)32 (40.5)
Air-disc brake configuration, n (%)307 (57.9)216 (58.2)46 (57.5)45 (57.0)
Drum/S-cam configuration, n (%)223 (42.1)155 (41.8)34 (42.5)34 (43.0)
Mean monthly distance, km12,840 ± 3,16012,910 ± 3,14012,670 ± 3,22012,700 ± 3,210
Baseline tire age, months14.2 ± 6.814.1 ± 6.714.4 ± 6.914.5 ± 7.0
Mountainous-route exposure, n (%)185 (34.9)130 (35.0)28 (35.0)27 (34.2)

The three vehicle-level data partitions were similar with respect to truck age, baseline mileage, duty cycle, braking configuration, monthly distance and mountainous-route exposure. Of the 82 hypothetical defect-positive trucks, 58 occurred in the development cohort, 11 in the validation cohort and 13 in the locked test cohort. The most frequent adjudicated brake abnormalities were thermal asymmetry/overheating and air-pressure response abnormalities, whereas progressive under-inflation/leak, low tread and visible structural tire damage were the most frequent tire findings.

Table 7: Diagnostic performance of Brake Tire Sense-AI in the locked test set (n=79)

MetricBrake defectsTire defectsCombined primary endpoint
Sensitivity, % (95% CI)85.7 (48.7-97.4)100.0 (64.6-100.0)92.3 (66.7-98.6)
Specificity, % (95% CI)94.4 (86.6-97.8)95.8 (88.5-98.6)92.4 (83.5-96.7)
PPV, % (95% CI)60.0 (31.3-83.2)70.0 (39.7-89.2)70.6 (46.9-86.7)
NPV, % (95% CI)98.6 (92.2-99.7)100.0 (94.7-100.0)98.4 (91.4-99.7)
AUROC (95% CI)0.94 (0.86-0.99)0.96 (0.90-1.00)0.95 (0.89-0.99)
Precision-recall AUC0.760.810.83
Brier score0.0840.0630.071
Median warning lead time, h96 (54-151)128 (72-194)112 (61-178)
Median warning lead distance, km356 (174-604)472 (233-748)418 (205-690)

In the hypothetical locked test set, 13 of 79 trucks (16.5%) experienced the combined primary endpoint. Brake Tire Sense-AI correctly identified 12 of these 13 trucks and generated five vehicle-level false-positive classifications within the pre-specified prediction horizon. This yielded sensitivity of 92.3% (95% CI 66.7-98.6%), specificity of 92.4% (83.5-96.7%), PPV of 70.6% (46.9-86.7%) and NPV of 98.4% (91.4-99.7%). Discrimination was high (AUROC 0.95, hypothetical 95% CI 0.89-0.99), with a precision-recall AUC of 0.83 and Brier score of 0.071. Median warning lead time for true-positive combined events was 112 hours (IQR 61-178), corresponding to 418 km (IQR 205-690) before mechanical confirmation.


Int J Engg Mgmt Res 2026;16(4)83
Naseer U, et al. Brake Tire Sense-AI

ijemr_1937_04.PNG
Figure 4: Receiver-operating-characteristic curve for the combined primary endpoint in the locked test set. Simulated values are shown for illustration only

Table 8: Exploratory operational outcomes in a matched conventional versus AI-assisted annual period

OutcomeConventional monitoring period/groupBrakeTireSense-AI period/groupEffect estimate (95% CI)
Confirmed major brake/tire defects7682Rate ratio 1.08 (0.79-1.47)
Roadside brake/tire breakdowns2412Rate ratio 0.50 (0.25-1.00)
Unplanned maintenance visits6139Rate ratio 0.64 (0.43-0.96)
Maintenance-related downtime, h/10,000 km7.94.8Mean difference -3.1 (-4.5 to -1.7)
False alerts / 10,000 kmNot applicable0.06Descriptive only
Near-miss events attributed to brake/tire condition157Rate ratio 0.47 (0.19-1.14)
Brake/tire-related crashes31Exploratory only

Compared with conventional monitoring, the hypothetical AI-assisted period showed fewer roadside brake/tire breakdowns (12 vs 24; rate ratio 0.50, 95% CI 0.25-1.00), fewer unplanned maintenance visits (39 vs 61; rate ratio 0.64, 95% CI 0.43-0.96), and lower maintenance-related downtime (4.8 vs 7.9 h/10,000 km; mean difference -3.1 h/10,000 km, hypothetical 95% CI -4.5 to -1.7). The AI false-alert burden was 0.06 per 10,000 km. Near-miss events were numerically lower (7 vs 15; rate ratio 0.47, 95% CI 0.19-1.14), while the very small number of brake/tire-related crashes (1 vs 3) precluded meaningful inference.

These operational comparisons are simulated examples and are not evidence of causal crash reduction.

ijemr_1937_05.PNG
Figure 5: Annual operational outcomes under conventional monitoring and BrakeTireSense-AI-assisted monitoring. Simulated counts are shown only to demonstrate reporting format

5.1. Engineering Interpretation of the Cross-Country Deployment Model

The central engineering value of Brake Tire Sense-AI is the transformation of maintenance from a periodic inspection problem into a continuous state-estimation problem. For each truck, the system maintains a dynamic health state for each wheel end and tire position. Measurements are normalized against vehicle load, speed, road gradient, ambient temperature and recent braking demand. This reduces false alarms caused by expected thermal or pressure changes and increases the probability that persistent deviations are interpreted as degradation.

The Pakistan and USA contexts also demonstrate why a single fixed threshold is unlikely to be optimal. A tire-pressure or brake-temperature threshold calibrated in one climate or duty cycle may generate excessive warnings or miss deterioration in another. A more robust engineering implementation should therefore use a hierarchical model consisting of hard safety limits, vehicle-specific baselines and a learned degradation model. Hard limits provide deterministic protection, while the AI layer detects sub-threshold changes and predicts progression.

At fleet scale, the system can be integrated with computerized maintenance-management systems. A high-risk prediction should automatically create a maintenance work order, identify the affected axle or wheel position, preserve the supporting sensor evidence and record the technician's inspection result.


Int J Engg Mgmt Res 2026;16(4)84
Naseer U, et al. Brake Tire Sense-AI

This closed-loop architecture allows the model to learn from confirmed faults and reduces the gap between algorithmic prediction and physical engineering action.

6. Discussion

Using the illustrative values presented for manuscript development, Brake Tire Sense-AI achieved high diagnostic sensitivity and negative predictive value for the combined brake/tire endpoint, with strong simulated discrimination. From an engineering perspective, the more important metric is not model accuracy alone but whether a prediction occurs sufficiently early to permit a safe maintenance intervention. The simulated warning lead time is therefore presented as an engineering-operational metric requiring validation under real vehicle loads, gradients, temperatures and maintenance cycles.

The study should specifically compare multimodal fusion with single-modality models. Brake temperature alone can be physiologically analogous to a nonspecific biomarker: it may rise normally during high-load downhill operation. Contextual interpretation is therefore essential. Persistent temperature asymmetry, prolonged cooling time, abnormal pressure response and vibration change occurring together would be more mechanically concerning than any one parameter in isolation. Similar reasoning applies to tires, where pressure change must be interpreted alongside temperature, wheel dynamics, tread measurements and visible structural condition.

The completed manuscript should also discuss external validity. Models trained on one manufacturer, tire specification, climate or fleet may not generalize. Prospective evaluation across several sites and vehicle-level data partitioning are intended to reduce this risk, but additional external validation will still be necessary. Federated approaches may later permit multi-fleet model improvement while limiting centralized raw-data sharing [26].

Ability to explain should be evaluated as a practical maintenance tool rather than only as an algorithmic feature. SHAP and local explanation methods can identify the signals responsible for a prediction [20,21], but attribution does not prove causality. Mechanical inspection must remain the decision-making reference during validation.

The engineering pathway from component diagnosis to injury and fatality reduction should also be explicitly separated. A diagnostic model can demonstrate detection performance and warning lead time, whereas reduction in crashes, serious injuries and deaths requires evidence that warnings trigger timely corrective maintenance and that corrected vehicles have fewer hazardous failures. Therefore, the proposed system should be evaluated as a safety intervention in a subsequent large-scale controlled or quasi-experimental fleet study.

7. Limitations

Several limitations should be considered when interpreting the proposed findings. The expected 15% defect-positive prevalence used for sample-size planning may differ across fleets and jurisdictions because of variations in vehicle age, operating conditions, maintenance practices, road environments, and regulatory inspection systems. Maintenance practices and inspection quality may also vary between participating sites despite the use of standardized operational definitions and diagnostic criteria. Furthermore, rare catastrophic brake and tire failures may remain insufficiently frequent to provide precise event-specific estimates, particularly during the initial validation period. Additional sensing hardware may be susceptible to sensor failure, calibration drift, contamination, vibration, moisture, dust, temperature extremes, and other environmental exposures encountered during commercial truck operation.

Similarly, computer-vision performance may vary according to illumination, tire cleanliness, mud or road debris, tire design, camera positioning, image resolution, and image-acquisition geometry. The predictive performance of the artificial intelligence models may also degrade following changes in vehicle components, tire or brake specifications, electronic control-unit firmware, sensor configurations, maintenance procedures, or fleet operating patterns, necessitating continuous model-drift surveillance, periodic recalibration, and prospective performance monitoring. Finally, the proposed study is primarily powered to evaluate diagnostic performance and early-warning capability rather than to provide statistically definitive evidence of reductions in fatal or severe crashes.


Int J Engg Mgmt Res 2026;16(4)85
Naseer U, et al. Brake Tire Sense-AI

Consequently, any observed reduction in crashes, serious injuries, or fatalities should initially be considered exploratory and hypothesis-generating until confirmed through larger, adequately powered longitudinal or multicenter interventional studies.

8. Ethical, Regulatory and Data Considerations

The study involves vehicle telemetry and operational data rather than human biomedical intervention; nevertheless, local institutional or organizational review should determine whether human-subject considerations arise from driver-linked data. Driver identifiers should not be used for performance evaluation unless separately approved. The research system must not override the driver or vehicle braking system during validation. Fleet safety procedures and mandatory maintenance requirements supersede study procedures at all times.

9. Conclusion

Brake Tire Sense-AI is an engineering-oriented, multimodal cyber-physical diagnostic framework for early detection of brake and tire degradation in heavy commercial trucks. Its design integrates physical sensing, vehicle-network data, edge computing, machine learning, computer vision, sensor fusion, explainable risk scoring and predictive maintenance. The Pakistan and USA road-safety contexts demonstrate the need for a system that can adapt to differences in road environment, vehicle configuration, climate, loading and maintenance practice. The illustrative results demonstrate how diagnostic accuracy and operational performance can be reported, but they are not a substitute for prospective empirical data. If real-world validation demonstrates reliable early detection and fleet operators consistently act on high-risk warnings, Brake Tire Sense-AI could reduce preventable brake- and tire-related mechanical failures, severe highway crashes, serious injuries and fatalities. Accordingly, the long-term engineering objective is not merely higher diagnostic accuracy, but a measurable reduction in injury and fatality rates through earlier fault detection, timely maintenance and safer vehicle operation.

References

[1] Federal Motor Carrier Safety Administration. Large Truck and Bus Crash Facts 2022 [Internet]. Washington (DC): U.S. Department of Transportation; 2025 [cited 2026 Aug 8].

[2] Commercial Vehicle Safety Alliance. CVSA releases 2025 International Roadcheck results [Internet]. Washington (DC): CVSA; 2025 Oct 7 [cited 2026 Aug 8].

[3] Commercial Vehicle Safety Alliance. CVSA releases 2025 Brake Safety Week results [Internet]. Washington (DC): CVSA; 2025 Oct 27 [cited 2026 Aug 8].

[4] Commercial Vehicle Safety Alliance. Nearly 600 commercial motor vehicles removed from North American roadways in one day due to brake violations [Internet]. Washington (DC): CVSA; 2026 May 19 [cited 2026 Aug 8].

[5] National Highway Traffic Safety Administration. Tire safety ratings and awareness: TireWise [Internet]. Washington (DC): U.S. Department of Transportation [cited 2026 Aug 8].

[6] Sepriyanto, Sumarsono DA, & Adhitya M, Sholahudin. (2026). Brake failure and fault detection in heavy vehicles: trends, challenges, and research gap from a systematic literature review. Teknosains J Sains Teknol Inform., 13(2), 400-411. doi:10.37373/tekno.v13i2.2164.

[7] Jain M, Vasdev D, Pal K, & Sharma V. (2022). Systematic literature review on predictive maintenance of vehicles and diagnosis of vehicle health using machine learning techniques. Comput Intell., 38(6), 1990-2008. doi:10.1111/coin.12553.

[8] Mahale Y, Kolhar S, & More AS. (2025). A comprehensive review on artificial intelligence driven predictive maintenance in vehicles: technologies, challenges and future research directions. Discov Appl Sci., 7, 243. doi:10.1007/s42452-025-06681-3.

[9] Lin SL. (2023). Research on tire crack detection using image deep learning method. Sci Rep., 13, 8027. doi:10.1038/s41598-023-35227-z.

[10] Saleh RAA, Konyar MZ, Kaplan K, & Ertunç HM. (2024). End-to-end tire defect detection model based on transfer learning techniques. Neural Comput Appl., 36, 12483-12503. doi:10.1007/s00521-024-09664-4.


Int J Engg Mgmt Res 2026;16(4)86
Naseer U, et al. Brake Tire Sense-AI

[11] Saleh RAA, & Ertunç HM. (2025). Attention-based deep learning for tire defect detection: fusing local and global features in an industrial case study. Expert Syst Appl., 269, 126473. doi:10.1016/j.eswa.2025.126473.

[12] Xie M, Bian H, Jiang C, Zheng Z, & Wang W. (2024). An improved YOLOv5 algorithm for tyre defect detection. Electronics, 13(11), 2207. doi:10.3390/electronics13112207.

[13] Chen J, Li A, Wang T, & Wang X. (2024). Research on tire internal defect identification method based on deep learning. Mech Eng Adv., 2(2), 1495. doi:10.59400/mea.v2i2.1495.

[14] Karkaria V, Chen J, Luey C, Siuta C, Lim D, & Radulescu R, et al. (2025). A digital twin framework utilizing machine learning for robust predictive maintenance: enhancing tire health monitoring. J Comput Inf Sci Eng., 25(7), 071003. doi:10.1115/1.4067270.

[15] Breiman L. (2001). Random forests. Mach Learn., 45(1), 5-32.

[16] Cortes C, & Vapnik V. (1995). Support-vector networks. Mach Learn., 20, 273-297. doi:10.1007/BF00994018.

[17] Hochreiter S, & Schmidhuber J. (1997). Long short-term memory. Neural Comput., 9(8), 1735-1780. doi:10.1162/neco.1997.9.8.1735.

[18] He K, Zhang X, Ren S, & Sun J. (2016). Deep residual learning for image recognition. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, p. 770-778. doi:10.1109/CVPR.2016.90.

[19] Redmon J, Divvala S, Girshick R, & Farhadi A. (2016). You only look once: unified, real-time object detection. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, p. 779-788.

[20] Lundberg SM, & Lee SI. (2017). A unified approach to interpreting model predictions. In: Guyon I, von Luxburg U, Bengio S, Wallach H, Fergus R, Vishwanathan S, et al., editors. Advances in Neural Information Processing Systems 30. Red Hook (NY): Curran Associates.

[21] Ribeiro MT, Singh S, & Guestrin C. (2016). Why should I trust you? Explaining the predictions of any classifier. In: Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, p. 1135-1144.

[22] Fayyad J, Jaradat MA, Gruyer D, & Najjaran H. (2020). Deep learning sensor fusion for autonomous vehicle perception and localization: a review. Sensors (Basel), 20(15), 4220. doi:10.3390/s20154220.

[23] Buderer NMF. (1996). Statistical methodology: I. Incorporating the prevalence of disease into the sample size calculation for sensitivity and specificity. Acad Emerg Med., 3(9), 895-900. doi:10.1111/j.1553-2712.1996.tb03538.x.

[24] Imteyaz S, & Iqbal S. (2026). Sensorless tire health monitoring system based on machine learning for passenger vehicles. SAE Technical Paper 2026-26-0670. Warrendale (PA): SAE International. doi:10.4271/2026-26-0670.

[25] Eddy CW, Castanier MP, & Wagner JR. (2025). Predictive maintenance of a ground vehicle using digital twin technology. SAE Int J Adv Curr Pract Mobil., 7(2), 865-876. doi:10.4271/2024-01-2867.

[26] von Wahl L, Heidenreich N, Mitra P, Nolting M, & Tempelmeier N. (2024). Data disparity and temporal unavailability aware asynchronous federated learning for predictive maintenance on transportation fleets. Proc AAAI Conf Artif Intell., 38(14), 15420-15428. doi:10.1609/aaai.v38i14.29467.

[27] Sharma J, Mittal ML, Soni G, & Keprate A. (2024). Explainable artificial intelligence approaches in predictive maintenance: A review. Recent Pat Eng., 18(5), e170423215860. doi:10.2174/1872212118666230417084231.

[28] International Organization for Standardization. ISO 26262-1:2018. Road vehicles - Functional safety - Part 1: Vocabulary. 2nd ed. Geneva: ISO; 2018.

[29] International Organization for Standardization; SAE International. ISO/SAE 21434:2021. Road vehicles - Cybersecurity engineering. Geneva: ISO; 2021.


Int J Engg Mgmt Res 2026;16(4)87
Naseer U, et al. Brake Tire Sense-AI

[30] SAE International. SAE J1939_202603: Serial control and communications heavy-duty vehicle network - Top-level document. Warrendale (PA): SAE International; 2026. doi:10.4271/J1939_202603.

[31] Khan A, et al. Which factors contribute to road crashes in non-commercial and commercial vehicles? An examination of administrative data from motorways in Pakistan. Traffic Inj Prev. 2021;22(7):541-546. doi:10.1080/17457300.2021.1972319.

[32] World Health Organization. Road safety Pakistan 2023 country profile [Internet]. Geneva: WHO; 2024 [cited 2026 Aug 11].

[33] Federal Motor Carrier Safety Administration. Crash statistics: large truck and bus crash facts [Internet]. Washington (DC): U.S. Department of Transportation; 2026 [cited 2026 Aug 11].

Disclaimer / Publisher's Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of Journals and/or the editor(s). Journals and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Int J Engg Mgmt Res 2026;16(4)88