All Categories

News

Home >  News

Comparison of filtration performance of commercially available automotive cabin air filters against various airborne pollutants

Time : 2024-11-06

Comparison of filtration performance of commercially available automotive cabin air filters against various airborne pollutants

       

Abstract

Automotive engine intake filters ensure clean air delivery to the engine, though over time these filters load with contaminants hindering free airflow. Today’s open-loop approach to air filter maintenance has drivers replace elements at predetermined service intervals, causing costly and potentially harmful over- and under-replacement. The result is that many vehicles consistently operate with reduced power, increased fuel consumption, or excessive particulate-related wear which may harm the catalyst or damage machined engine surfaces.
We present a method of detecting filter contaminant loading from audio data collected by a smartphone and a stand microphone. Our machine learning approach to filter supervision uses Mel-Cepstrum, Fourier and Wavelet features as input into a classification model and applies feature ranking to select the best-differentiating features. We demonstrate the robustness of our technique by showing its efficacy for two vehicle types and different microphones, finding a best result of 79.7% accuracy when classifying a filter into three loading states.
Refinements to this technique will help drivers supervise their filters and aid in optimally timing their replacement. This will result in an improvement in vehicle performance, efficiency, and reliability, while reducing the cost of maintenance to vehicle owners.

Introduction

Every year, the average vehicle’s age and annual miles traveled increase United States Department of Transportation (2016b), IHS Inc (2016) and with the shift toward shared mobility, the need for efficient, reliable and durable vehicles continues to grow.
Most of the 260-million vehicle U.S. light-duty fleet (United States Department of Transportation, 2016a) is gasoline powered (U.S. Energy Information Administration, 2016), with engines that consume air and fuel, ignite this mix to propel a piston, and exhaust combustion byproducts. Any inefficiency causes engine performance, economy and longevity to suffer.
The intake system is critical to optimal performance. Incoming air must be free-flowing to attain efficiency, clean, to protect engine surfaces against abrasion and cold, so that the increased density allows more fuel to be combusted, improving power.
A key element of engine intakes, filters reduce contaminant concentration to safe levels (Jaroszczyk et al., 1993) while ensuring free fluid flow to limit intake air heating. These filters are wear items, needing cleaning or replacement once loaded with dirt, dust, and debris.
Optimal filtration improves particulate entrapment, reducing engine cylinder erosion. Small changes to efficiency have significant impact: engine wear is 8 times faster for a filter that is 98% versus 99% efficient (Jaroszczyk et al., 1993). Further, ideal filtration reduces cabin noise levels and improves engine power and response. In contrast, dirty filters limit power, cause noise, waste fuel Norman et al. (2009), Toma (2016) and may cause downstream catalytic converter failures. These challenges are most significant in carbureted vehicles (Thomas et al., 2012) lacking closed-loop fuel control. While new cars switched to fuel injection by the mid-1990s, many cars, motorcycles and other light transport vehicles around the world still use carburetors.
Changing filter elements early seems an obvious solution, but early replacement causes subtle but serious problems. Particulate capture efficiency increases with loading (Norman et al., 2009), so lightly used elements reduce engine wear and extend service life. There exists an optimal window in which to change a filter — one in which the filter captures a majority of particulates and minimally restricts flow.
In-vehicle sensors have been designed to solve the problem of optimal filter replacement, but most new vehicles with On-Board Diagnostics typically do not monitor this condition and older vehicles typically lack any sensing. Few vehicles offer vacuum-based intake pressure drop sensors (Norman et al., 2009) that indicate an increase over baseline pressure drop of 1-2.5kPa Toma (2016), Thomas et al. (2012). Where sensors are not present, drivers are typically unaware and therefore rely on data-blind timing, with most drivers replacing filters at set intervals (often 15,000 km (Toma, 2016)) or when they look dirty. These methods are inaccurate, with vehicles used in varied environments with different particulate loads and unpredictable airflow rates (Jaroszczyk et al., 1993).
In a survey of 21 air filters tested after removal, 15 were removed early while two had been changed after performance-degrading occlusion begun (Toma, 2016). This indicates that drivers taking vehicles in for service change filters too early but is inherently biased, as the dirtiest filters are found in those cars never taken for service. Assuming a 2% loss in fuel economy in the under-serviced vehicles, an average driver spending $1680 per year in fuel wastes $33.60 driving with a dirty filter. This exceeds a typical filter’s cost and demonstrates the potential savings for optimal replacement timing, not to mention the long term damage to engines and catalytic convertors.
Streamlined, realtime filter classification could reduce vehicle operating costs and emissions while improving reliability. There exists latent demand for this information — 81.4%of people would take recommendations from a data-informed system (Toma and Bobalca, 2016).
To reduce the need for behavioral changes, low-cost, pervasive sensing using smartphones may be used to repurpose existing devices (Engelbrecht et al., 2015). In recent years, consumer electronics manufacturers have increased mobile sensing capabilities. These new inputs, ranging from atmospheric pressure and device orientation to temperature, touch, and proximity, have met with commensurate enhancements in mobile computation, storage, and connectivity (Han and Cho, 2016). Our own work has shown that it is possible to monitor engine ignition using such devices (Siegel et al., 2016b).
We aim to transition from today’s reactive maintenance paradigm to proactive, availing ourselves of these resources. We apply mobile audio to observe how a car “breathes” to classify air filter performance with the goal of creating a “remaining life” indicator and condition monitor for air filters to enhance compliance with automotive best maintenance practices. This paper demonstrates how mobile audio data and ensemble classification may be applied to categorizing air filter condition into multiple loaded states.
In Section 2, we hypothesize that sound emanating from the intake changes with particulate loading, while Section 3 explores related work. Section 4 describes an experimental procedure to collect data and simulate contaminants restricting airflow. We describe our ensemble classification algorithm in Section 5 and present results in Section 6, showing high accuracy in differentiating new, gently used, dirty and obstructed filters. Finally, Section 7 discusses of future improvements for this algorithm and applications of pervasive sensing to other vehicle faults.

Section snippets

Problem description

The ideal combustion engine demands a limitless supply of free-flowing, clean, cold air. In reality, engines require filters to clean air and limit wear. When new, these filters restrict intake airflow, and as the filter loads with contaminants, this restriction and related pressure drop increase. While intake systems are tuned to minimize noise, vibration and harshness, changes in flow ultimately lead to perceptible changes in the audio emanating from the intake. We assert that these pressure

Prior art

Characterizing vehicle performance, classifying component condition and identifying abnormal behavior using time-domain signals is not a new field. In-vehicle sensing in particular has been applied to air filter monitoring. However, multi-state classification and pervasive condition monitoring remain underexplored.
This is not for lack of pervasive vehicle diagnostics. The use of audio signals is especially prevalent in research and industrial applications because acoustic signals do not require 

Experimental procedure and hypothesis validation

This section describes how we generated audio samples from a vehicle with varying degrees of air filter contamination to train a three-state classifier for multiple vehicles.
We first discuss an experiment collecting data to prove the concept of using audio features to differentiate old from new filters. Then, we present a procedure for generating controlled data for multi-state contamination classification and explain how this approach assures our classifiers’ robustness. Here, we collect data

Algorithm development

From Fig. 5, we hypothesized that the FT peak differences could be used to differentiate among three states, with additional features improving classification accuracy. In this section we discuss how we generated features, tuned a classifier, and selected the optimal input parameters to maximize filter loading classification accuracy while minimizing overfitting.

Out-Sample results

This section shows the outsample, optimized results for the three tested data sets in tabular form.
In Table 1, we see the optimal configuration for each of the three models and the 5-fold cross-validated insample performance as well as the 25% outsample performance.
We note trends in optimal model parameters. Each model tends to select small bin sizes, suggesting that the DFT elements will play an important role in differentiating states and that the features of interest are focused on narrow

Conclusions

We demonstrated 80% accuracy in three-state air filter particulate loading detection using MFCC, DFT and wavelet features and bagged decision trees, proving the viability of batch processed smartphone audio for filter classification. Multi-state classification is a step towards condition monitoring, while the demonstrated classifier’s sensitivity suggests early response is possible. A mobile application using this approach may ultimately improve vehicle performance and efficiency.
The results

PREV : THE VITAL IMPORTANCE OF THE OIL FILTER

NEXT : The Real Story on Automotive Filters