Machine Learning Predicts Emissions of Brake Wear PM_2.5: Model Construction and Interpretation

Abstract

Brake emissions are generated every time a brake is applied to a vehicle. However, revealing the pattern of brake emissions under different operating conditions is conventionally considered highly challenging. Here, we compiled a brake wear PM2.5 data set collected from brake dynamometer simulation experiments and obtained the mapping relationship between brake emissions and influencing factors through a machine learning (ML) method. The random forest model was devised and displayed good prediction performance with an R 2 of 0.89 on the test set. Model-related (similarity network analysis) and model-unrelated (partial dependence plots and centered-individual conditional expectation plots) interpretation methods were used to break the black box of ML to obtain the marginal contribution of the model input feature parameters (brake energy dissipation, average temperature during braking, brake pad metal content, and brake pad surface area) to the model output results. This study suggests that avoiding rapid braking behavior and using brake pads with a lower metal content are feasible ways to reduce brake wear PM2.5 emissions. The development of a ML-based brake emission model provides novel insights into the accurate assessment and control of brake emissions.

Machine Learning Reveals the Parameters Affecting the Gaseous Sulfuric Acid Distribution in a Coastal City: Model Construction and Interpretation

“…24,25 RF models have been successfully used in forecasting studies of environmental pollutants, such as estimating the emission of brake wear in PM 2.5 and predicting the OH radicals near the surface in North American cities. 26,27 In this study, an RF model was applied to simulate the SA concentration and to examine the impacts of parameters. To clarify the influences of single-input and multi-input feature parameters on the model outcomes, two black-box visualization tools, partial dependence plots (PDPs) and centered-individual conditional expectation (c-ICE), were carefully scrutinized.…”

“…Despite the multiple SA formation mechanisms, machine learning (ML) is a data-driven strategy that could minimize the relationships among complex data. , A decision tree-based integrated supervised ML system called random forest (RF) helps identify nonlinear correlations between parameters. , RF models have been successfully used in forecasting studies of environmental pollutants, such as estimating the emission of brake wear in PM 2.5 and predicting the OH radicals near the surface in North American cities. , In this study, an RF model was applied to simulate the SA concentration and to examine the impacts of parameters. To clarify the influences of single-input and multi-input feature parameters on the model outcomes, two black-box visualization tools, partial dependence plots (PDPs) and centered-individual conditional expectation (c-ICE), were carefully scrutinized. − The development of ML-based models provides new perspectives for comprehension of SA mechanisms.…”

Predicting Atmospheric Particle Phase State Using an Explainable Machine Learning Approach Based on Particle Rebound Measurements

“…For instance, PDPs were used to reveal the factors that influence human PM 2.5 exposure, 36 and a similarity network was established to predict the PM 2.5 mass concentrations emitted by vehicle tire wears. 37 Among these methods, the SHAP approach shows the best accuracy in explaining tree models owing to its high consistency. 38 The SHAP approach is developed based on coalitional game theory 34,35 and has been successfully implied in the field of atmospheric environment, such as identifying the driving factor of particulate matter air pollution.…”

“…Approaches such as partial dependency plots (PDPs), centered-individual conditional expectation (c-ICE), similarity network, and SHapley Additive exPlanation (SHAP) , have been used to explain the outcomes of ML models. For instance, PDPs were used to reveal the factors that influence human PM 2.5 exposure, and a similarity network was established to predict the PM 2.5 mass concentrations emitted by vehicle tire wears . Among these methods, the SHAP approach shows the best accuracy in explaining tree models owing to its high consistency .…”

New Insights on the Formation of Nucleation Mode Particles in a Coastal City Based on a Machine Learning Approach

“…While the studies highlighted earlier illuminate the PNCN determinants under diverse environmental contexts, identifying the key parameters that affect PNCN and investigating the marginal effects of individual (or multiple) feature variables on PNCN remain challenging. Machine learning (ML) has the potential to handle and learn from large, complex, and multidimensional data sets to develop simulative models. , ML methods, such as random forest (RF), support vector machine (SVM), and artificial neural network (ANN), have been applied to model pollutants in the atmosphere. − For instance, our research team utilized a similar ML approach with data from the same site to develop a model for simulating gaseous H 2 SO 4 concentrations and assess the influence of different feature parameters on the output variable . ML techniques adeptly capture intricate nonlinear relationships, paving the way for comprehensive interpretations.…”