Highly Accurate Tire Loss of Pressure Detection Based on k-omega Turbulence Model and Random Forest Algorithm

The safety and performance of automobiles are significantly influenced by tire conditions, particularly variations in pressure and temperature, which directly impact driving stability and fuel efficiency. Traditional direct tire pressure monitoring systems require additional sensors, which can be costly and complex to maintain. In contrast, existing indirect tire pressure monitoring methods have limitations in multiple aspects of data processing. On one hand, these methods typically rely solely on wheel speed signal analysis, which is susceptible to environmental factors, thereby reducing data accuracy. On the other hand, most indirect tire pressure monitoring methods determine the loss of pressure status by comparing the difference in wheel speeds between different tires, a method that is often difficult to accurately detect when a similar degree of loss of pressure occurs in all four wheels at the same time, and can easily lead to problems with delayed alarms. To this end, this paper proposes a method based on the k-omega turbulence model to achieve the simulation of the tire wall temperature field, and based on the Random Forest Algorithm for high-accuracy indirect tire loss of pressure detection. Vehicle signals are collected through the ABS's wheel speed sensors and the ECU. Coupled with computational fluid dynamics (CFD) simulations of tire temperature fields, the temperature gradient distributions of the tire walls are calculated. Decision tree algorithms classify the data, extracting turbulence kinetic energy and dissipation rate as primary features, while coordinates and temperature distributions serve as auxiliary features to construct a random forest model. Tire deflation levels are categorized into " brand-new", "normal", "dangerous" and " extremely dangerous" states, the degree of tire pressure loss was evaluated by the random forest model to classify and predict different tire pressure states. Experiments indicate that at speeds ranging from 40 to 80 km/h, the method achieves a prediction accuracy of 100% for tires in hazardous and critical deflation states. In all scenarios, the prediction accuracy exceeds that of Naive Bayes and logistic regression models by 10% to 40%, highlighting its superior practical value.