MAF nearly tuned
I mentioned before, I used this perl script to calibrate my MAF sensor. The script works on a simple principle: If, at a given frequency output, the MAF table over or underpredicted the air mass by a given percent, simply scale that entry in the table by the same percent. Seems simple enough, right?
The script itself scans a set of log files looking at the LTFT, STFT, MAF Frequency and commanded AFR in each row. It throws out rows for which the computer commands an AFR other than 14.7. The script then adds the LTFT and STFT together to get the final fuel trim, and drops it into one of 81 bins based on what the current MAF frequency is. At the end, it produces an average fuel trim for each bin and smooths the result with a weighted average of neighbors.
With this technique, there will be gaps-bins that have no data. For the left and right edge of the histogram, I took the conservative approach of merely replicating the nearest result out to the edge. For gaps in the middle, I just linearly interpolate across the gaps.
The results have been pretty good. As you can see, the last two iterations of tuning moved the parameters much less than 5% on average. (The vertical axis is "scale factor", so 1.1 means "Increase 10%.")
I seemed to have converged pretty quickly. Cool, eh? (For the record, being within +/-4% is considered "well tuned.") Next, I need to work on the correct extrapolation for the far right side of the graph, because I'm sure at higher flow rates I'm going to underpredict the airmass.
The script itself scans a set of log files looking at the LTFT, STFT, MAF Frequency and commanded AFR in each row. It throws out rows for which the computer commands an AFR other than 14.7. The script then adds the LTFT and STFT together to get the final fuel trim, and drops it into one of 81 bins based on what the current MAF frequency is. At the end, it produces an average fuel trim for each bin and smooths the result with a weighted average of neighbors.
With this technique, there will be gaps-bins that have no data. For the left and right edge of the histogram, I took the conservative approach of merely replicating the nearest result out to the edge. For gaps in the middle, I just linearly interpolate across the gaps.
The results have been pretty good. As you can see, the last two iterations of tuning moved the parameters much less than 5% on average. (The vertical axis is "scale factor", so 1.1 means "Increase 10%.")
I seemed to have converged pretty quickly. Cool, eh? (For the record, being within +/-4% is considered "well tuned.") Next, I need to work on the correct extrapolation for the far right side of the graph, because I'm sure at higher flow rates I'm going to underpredict the airmass.