SDEC Automates Calibration of Diesel Particulate Filter Soot Load Models
“Model-Based Calibration Toolbox helped us improve the quality of our calibrations, reduce calibration costs, and solve the problems of traditional manual calibration, which relies too heavily on the experience of engineers.”
Challenge
Reduce emissions and lower costs by developing a diesel particulate filter model that acts as a virtual sensor
Solution
Use Simulink and Model-Based Calibration Toolbox to create and then automatically calibrate an accurate model for diesel particulate filter soot accumulation
Results
- Calibration accuracy improved by 90%
- Consistent results achieved
- Calibration time shortened
Diesel-engine testing at Shanghai Diesel Engine Co. (SDEC).
Even as electric vehicle usage continues to ramp up, minimizing emissions from diesel trucks and other vehicles remains imperative for automotive manufacturers, particularly in China, where stringent Stage 6 emissions standards are being phased in. Diesel particulate filters (DPF) play a crucial role in managing emissions by removing soot from diesel exhaust.
To keep DPFs running at maximum efficiency, the electronic control unit (ECU) software uses a DPF soot loading model as a virtual sensor to estimate carbon deposit buildup and determine when the DPF needs to be cleaned through a process known as active regeneration. The calibration of this soot model is often a complicated manual process that is highly dependent on the expertise of specialized engineering teams.
Soot loading models are important because during active regeneration, extra fuel is passed through the engine to burn off the soot affecting overall vehicle fuel economy. Soot models that overestimate soot loading use more fuel than necessary, and soot models that underestimate soot loading can lead to accumulated soot in the DPF over the long run.
Engineers at Shanghai Diesel Engine Co. (SDEC) use MATLAB® and Model-Based Calibration Toolbox™ to automatically calibrate soot models that provide more accurate estimates than simpler and less capable manually calibrated models. “Model-Based Calibration Toolbox enabled us to improve calibration accuracy and meet engineering requirements for soot estimation,” says Xiujuan Xia, the senior engineer in charge of the DPF system for diesel engines at SDEC. “Just as important, it enabled us to shorten development cycles, avoid tedious manual tuning of our models, and implement workflows that don’t depend so heavily on the experience of the engineer.”
Challenge
SDEC engineers collect data for the calibration process via a series of on-road driving tests, during which many large transient datasets are recorded. For each test, the engineers weigh the DPF before and after driving the vehicle under a variety of conditions, including different engine speeds, torques, altitudes, and temperatures. All these data – including DPF weights and driving conditions – must be cleaned before being used to calibrate values in dozens of lookup tables in the soot load model.
In the past, SDEC calibrated the lookup table values manually, which made it difficult to get accurate results quickly. Errors in the model resulted in either underestimates of soot load, and thus inefficient filtering, or overestimates of soot load, which led to frequent regenerations and excess fuel consumption. The size of the datasets and large number of adjustable tables can also cause long transient calibration wait times for in-house optimizers.
As with many diesel engine companies in the heavy transportation and agriculture sector, the quality of calibration and results are largely a function of engineering experience and the right technology.
Solution
Working in Simulink®, SDEC engineers created a soot load model based on specifications and documentation provided by the OEM supplier of the DPF.
The team worked with MathWorks engineers to develop a MATLAB script for cleaning and preprocessing the vehicle road test data from 50 different data files.
Using the Model-Based Calibration Toolbox in MATLAB, the team then generated optimal calibrations for lookup tables in the soot load model based on the cleaned data. Inputs to the model included measurements for engine speed and torque as well as other operating conditions; the model’s output was soot accumulation as represented by the weight of the DPF. What’s more, SDEC can now handle large transient datasets and many tables efficiently through the specialized optimization techniques in Model-Based Calibration Toolbox, resulting in better results delivered in much less time than previous methods.
To further reduce calibration times, the team used Parallel Computing Toolbox™ to complete the optimizations on a multicore workstation.
The engineers visualized the calibration results using MATLAB to generate plots of the deviation between the model’s soot load estimates in grams per liter (g/L) and the actual measured values from the drive-cycle testing.
The calibrated lookup table values were then incorporated into an ECU, which is now in use on production vehicles.
SDEC engineers collect data for the calibration process via a series of on-road driving tests.
Results
- Calibration accuracy improved by 90%. “With our manual process, we often had difficulty achieving our baseline accuracy requirements, which is an error of less than 1 g/L for soot accumulation,” says Xia. “Using Model-Based Calibration Toolbox, we improved accuracy by 90% and reduced the error to below 0.1 g/L.”
- Consistent results achieved. “Engineers with the necessary expertise to perform complex calibrations are in short supply, and as a result it’s difficult to get consistent results,” says Xia. “With Model-Based Calibration Toolbox, we consistently get the same high-quality result, even when a junior engineer is performing the calibration.”
- Calibration time shortened. “When we tuned calibration lookup tables in the past, it took weeks to achieve the level of accuracy we needed, especially if transients were involved,” notes Xia. “We estimate that automating the process with Model-Based Calibration has shortened calibration time from weeks to hours.”
