Page organisation in this revision
- Page type
- Technical article
- Subject
- Tuning
- Parent
- Section level
- Applies to
- No ECU-specific applicability
- Tool groups
- None
- Manufacturers
- None
Changes from revision 1 to revision 1
Removed text is struck through; added text is highlighted.
The ECU is allowed to remember
A map is the starting estimate. Several of the loops around it are allowed to store a correction so the next trip begins closer to the engine they last saw. Bosch's diesel fuel-quantity mean-value adaptation is one documented case. It runs in the lower part-load range and works out the average difference between the injected quantity the ECU asked for and the quantity implied by the lambda sensor and the air-mass meter. The differences are stored at learning points in EEPROM and are available as soon as the engine is next started. In the indirect mode those corrections feed the setpoint maps for EGR, boost and start of injection. In the direct mode they also correct the metered quantity. Bosch, fuel-quantity mean-value adaptation, printed pages 58–62.
That is why a diesel can feel different after a tank of fuel, an air leak, or a period of short trips, without anybody editing a file. The stored correction was built in one region. It is then applied, without waiting, when the operating point next passes a learned site. A part-load smoke or timing change that appears only after the engine has been driven, and that survives a restart, fits this pattern better than it fits a pedal map.
The same booklet does not claim this lambda adaptation for every EDC. Later ECUs have their own learning functions. Treat the Bosch description as the principle, and confirm the function on the software in front of you.
Petrol trims are the same kind of memory
ME7 separates a short-term mixture correction, which reacts while the loop is running, from long-term adaptations stored by region. Additive and multiplicative terms, and each bank, are different corrections. The S4 notes say to get the injector and airflow calibration right before using idle trim to chase injector latency, because a later change to the base calibration moves the apparent latency error. They also say to keep the adaptation state with the log, so a learned change is not mistaken for the map you just edited. ME7 fuelling and MAF calibration.
ME7 boost control has its own learned integral correction. The same requested pressure can produce a different wastegate duty on two days because the adaptation has moved. ME7 torque, load and boost. Idle, throttle and gearbox learning on other families follow the same rule: read the learned value before you edit the base table it is correcting.
What drift feels like
Fine when cold, worse after a few miles, or the reverse, can be a region that learns only once the engine is in closed loop. Cold start itself is often open-loop. See cold start and warm-up.
A correction that grows at idle and then spoils the pull-away is a low-load learning region being asked to represent a different injector or airflow error. Fix the base model in the region where feedback exists.
Boost that hunts more each drive can be an integral adaptation winding up against a leak, a sticky actuator or a target the turbo cannot hold. See turbo response.
Two files that look identical and drive differently may not have the same EEPROM. A map compare does not show the learned cells.
How to use a reset
Clearing adaptations returns the ECU to the base calibration. That is useful as a test. It is not a repair if the base calibration, or the hardware, is what the adaptation was covering. After a reset, log the correction as it learns again, in the same operating region as the complaint. If the correction walks back to the same value, the engine is still asking for it. If the drivability complaint disappears and the correction stays small, the stored value was the stale part.
Record the adaptation values in the log that judges a calibration change. Otherwise the result you attribute to a map edit may be the ECU re-learning around it.
Sources
Bosch Electronic Diesel Control (EDC), 2001, independent PDF mirror. Fuel-quantity mean-value adaptation and its effect on EGR, boost and start-of-injection setpoints. Printed pages 58–62. Accessed 2026-11-11.
ME7 fuelling and MAF calibration and ME7 torque, load and boost. Short-term and long-term fuel corrections, and boost integral adaptation, as documented for that family.
Rendered revisions
Open each version to inspect formatting, tables and photographs.
Revision 1: Adaptations and calibration drift
The ECU is allowed to remember
A map is the starting estimate. Several of the loops around it are allowed to store a correction so the next trip begins closer to the engine they last saw. Bosch's diesel fuel-quantity mean-value adaptation is one documented case. It runs in the lower part-load range and works out the average difference between the injected quantity the ECU asked for and the quantity implied by the lambda sensor and the air-mass meter. The differences are stored at learning points in EEPROM and are available as soon as the engine is next started. In the indirect mode those corrections feed the setpoint maps for EGR, boost and start of injection. In the direct mode they also correct the metered quantity. Bosch, fuel-quantity mean-value adaptation, printed pages 58–62.
That is why a diesel can feel different after a tank of fuel, an air leak, or a period of short trips, without anybody editing a file. The stored correction was built in one region. It is then applied, without waiting, when the operating point next passes a learned site. A part-load smoke or timing change that appears only after the engine has been driven, and that survives a restart, fits this pattern better than it fits a pedal map.
The same booklet does not claim this lambda adaptation for every EDC. Later ECUs have their own learning functions. Treat the Bosch description as the principle, and confirm the function on the software in front of you.
Petrol trims are the same kind of memory
ME7 separates a short-term mixture correction, which reacts while the loop is running, from long-term adaptations stored by region. Additive and multiplicative terms, and each bank, are different corrections. The S4 notes say to get the injector and airflow calibration right before using idle trim to chase injector latency, because a later change to the base calibration moves the apparent latency error. They also say to keep the adaptation state with the log, so a learned change is not mistaken for the map you just edited. ME7 fuelling and MAF calibration.
ME7 boost control has its own learned integral correction. The same requested pressure can produce a different wastegate duty on two days because the adaptation has moved. ME7 torque, load and boost. Idle, throttle and gearbox learning on other families follow the same rule: read the learned value before you edit the base table it is correcting.
What drift feels like
- Fine when cold, worse after a few miles, or the reverse, can be a region that learns only once the engine is in closed loop. Cold start itself is often open-loop. See cold start and warm-up.
- A correction that grows at idle and then spoils the pull-away is a low-load learning region being asked to represent a different injector or airflow error. Fix the base model in the region where feedback exists.
- Boost that hunts more each drive can be an integral adaptation winding up against a leak, a sticky actuator or a target the turbo cannot hold. See turbo response.
- Two files that look identical and drive differently may not have the same EEPROM. A map compare does not show the learned cells.
How to use a reset
Clearing adaptations returns the ECU to the base calibration. That is useful as a test. It is not a repair if the base calibration, or the hardware, is what the adaptation was covering. After a reset, log the correction as it learns again, in the same operating region as the complaint. If the correction walks back to the same value, the engine is still asking for it. If the drivability complaint disappears and the correction stays small, the stored value was the stale part.
Record the adaptation values in the log that judges a calibration change. Otherwise the result you attribute to a map edit may be the ECU re-learning around it.
Sources
Revision 1: Adaptations and calibration drift
The ECU is allowed to remember
A map is the starting estimate. Several of the loops around it are allowed to store a correction so the next trip begins closer to the engine they last saw. Bosch's diesel fuel-quantity mean-value adaptation is one documented case. It runs in the lower part-load range and works out the average difference between the injected quantity the ECU asked for and the quantity implied by the lambda sensor and the air-mass meter. The differences are stored at learning points in EEPROM and are available as soon as the engine is next started. In the indirect mode those corrections feed the setpoint maps for EGR, boost and start of injection. In the direct mode they also correct the metered quantity. Bosch, fuel-quantity mean-value adaptation, printed pages 58–62.
That is why a diesel can feel different after a tank of fuel, an air leak, or a period of short trips, without anybody editing a file. The stored correction was built in one region. It is then applied, without waiting, when the operating point next passes a learned site. A part-load smoke or timing change that appears only after the engine has been driven, and that survives a restart, fits this pattern better than it fits a pedal map.
The same booklet does not claim this lambda adaptation for every EDC. Later ECUs have their own learning functions. Treat the Bosch description as the principle, and confirm the function on the software in front of you.
Petrol trims are the same kind of memory
ME7 separates a short-term mixture correction, which reacts while the loop is running, from long-term adaptations stored by region. Additive and multiplicative terms, and each bank, are different corrections. The S4 notes say to get the injector and airflow calibration right before using idle trim to chase injector latency, because a later change to the base calibration moves the apparent latency error. They also say to keep the adaptation state with the log, so a learned change is not mistaken for the map you just edited. ME7 fuelling and MAF calibration.
ME7 boost control has its own learned integral correction. The same requested pressure can produce a different wastegate duty on two days because the adaptation has moved. ME7 torque, load and boost. Idle, throttle and gearbox learning on other families follow the same rule: read the learned value before you edit the base table it is correcting.
What drift feels like
- Fine when cold, worse after a few miles, or the reverse, can be a region that learns only once the engine is in closed loop. Cold start itself is often open-loop. See cold start and warm-up.
- A correction that grows at idle and then spoils the pull-away is a low-load learning region being asked to represent a different injector or airflow error. Fix the base model in the region where feedback exists.
- Boost that hunts more each drive can be an integral adaptation winding up against a leak, a sticky actuator or a target the turbo cannot hold. See turbo response.
- Two files that look identical and drive differently may not have the same EEPROM. A map compare does not show the learned cells.
How to use a reset
Clearing adaptations returns the ECU to the base calibration. That is useful as a test. It is not a repair if the base calibration, or the hardware, is what the adaptation was covering. After a reset, log the correction as it learns again, in the same operating region as the complaint. If the correction walks back to the same value, the engine is still asking for it. If the drivability complaint disappears and the correction stays small, the stored value was the stale part.
Record the adaptation values in the log that judges a calibration change. Otherwise the result you attribute to a map edit may be the ECU re-learning around it.
Sources