Automotive suppliers: control, tolerances and tests under cost

The production plan depends on one planner with a spreadsheet

What happens. The plan for presses, lines and shifts sits in a spreadsheet that one person understands. Orders change daily, and that person re-plans by feel and by memory. Everybody else uses the result without seeing how it came about.

What it costs. A bottleneck on one person, a risk with every sickness or holiday, and plans that are feasible but not good: too many changeovers, late deliveries, idle machines.

Why the usual fix fails. A planning system bought from a catalogue does not know the rules the planner carries in his head. Documenting the spreadsheet freezes it. Training a second person takes years of the same experience.

How a fixed-price remote diagnosis checks it. From the spreadsheet, order lists and past plans I write down the rules and see how good the current plans are against what is possible. The fixed-price remote diagnosis tells you whether an optimisation tool your team can run itself would reproduce the planner’s results, and what it would improve.

Background: Trajectory optimisation and setpoint generation

Customers demand tighter tolerances and higher process capability

What happens. Vehicle makers and medical customers ask for narrower tolerances and for proof that the process holds them over time. Existing processes met the old limits with some luck, and variation from material, temperature and tool wear now eats the margin.

What it costs. Rejects, sorting effort, customer complaints and contracts at risk when a capability figure falls short.

Why the usual fix fails. Tighter incoming checks move the problem and add cost. Operators correcting by hand add their own variation. A new machine is a large investment for a problem that may lie in the control.

How a fixed-price remote diagnosis checks it. From process and measurement data of past batches I split the variation into what comes from the material, from drift and from the control itself. The fixed-price remote diagnosis tells you whether model-based control that reacts to these causes can narrow the spread on your current machines.

Background: Industrial process predictive control

Cabin heating and battery cooling compete for the same energy

What happens. With a heat pump in an electric car, the cabin, the battery and the electronics all want heat or cooling at the same time. The thermal system has valves, pumps and several modes. Switching between them is by rules written for typical days.

What it costs. Range lost in winter and summer, battery temperatures outside their best window, and comfort complaints.

Why the usual fix fails. More rules for more situations make the logic hard to test. A bigger heat pump costs money and weight. Tuning to a test cycle does not hold in real traffic, weather and driver wishes.

How a fixed-price remote diagnosis checks it. From vehicle test data (temperatures, flows, compressor power, driving profile) and the thermal layout I look at where energy is wasted by the current logic. The fixed-price remote diagnosis tells you whether predictive energy management using route and weather can save range.

Background: Economic predictive control

More sensors per vehicle add cost

What happens. Each new function wants its own sensor: another temperature probe, another pressure sensor, another wire and plug. Over a million vehicles the cost and the failure points add up.

What it costs. Part cost, harness weight, assembly time, warranty cases from sensor and connector faults.

Why the usual fix fails. Dropping the sensor loses the function. Cheaper sensors are less accurate and drift. Every function asking for its own sensor is a design habit, not a physical need.

How a fixed-price remote diagnosis checks it. From vehicle data with and without the sensor I look at how well the quantity can be computed from signals that are already there. The fixed-price remote diagnosis tells you which sensors could be replaced by a computed value, and how accurate that value would be.

Background: Computed measurements

On-board diagnosis requirements are hard to meet

What happens. Diagnosis functions must detect faults in sensors, actuators and emissions systems reliably, without false alarms, and within the rules. Each new system adds more checks, and they interact.

What it costs. Late changes when a check triggers too often or too rarely, compliance risk, and workshop visits for faults that are not there.

Why the usual fix fails. Fixed thresholds work for new parts and not for aged ones. Tightening them brings false alarms, loosening them brings misses. Testing every fault case on the vehicle is slow and expensive.

How a fixed-price remote diagnosis checks it. From fault-injection tests and field data I look at how the diagnostic signals behave with and without faults and where thresholds overlap. The fixed-price remote diagnosis tells you whether model-based diagnosis separates faults from ageing better, and what it needs.

Control functions are specified loosely and reworked late

What happens. A control function arrives as a document in prose, with open points. The developers interpret it, the tests find the gaps, and the customer changes the wording. The architecture shifts while the software is being written.

What it costs. Rework late in the project, disagreements about what was agreed, and delays at every software release.

Why the usual fix fails. More review meetings catch some gaps and not the control-specific ones, such as stability or limits. A tool for requirements stores text without checking whether the control can meet it.

How a fixed-price remote diagnosis checks it. From the specification, interface lists and existing simulation models I look at which requirements can be checked early on the control loop and which are ambiguous. The fixed-price remote diagnosis tells you what a clearer specification and architecture would look like.

Functions move to central computers and hit real-time limits

What happens. Vehicle functions that once ran on separate control units are moved to a few central computers. They share computing time and communication, and a function that had its own processor now has a time budget.

What it costs. Control functions that miss their deadlines, simplified algorithms that perform worse, and integration problems found late.

Why the usual fix fails. A faster processor costs money and heat. Cutting the sampling rate hurts control quality. Running the unchanged code on the new platform shows the problems only at integration.

How a fixed-price remote diagnosis checks it. From the function code or its description, timing measurements and the target platform data I look at where the time goes and which parts can be made cheaper without losing quality. The fixed-price remote diagnosis tells you whether the control can fit its budget, and how.

Background: Predictive control on embedded hardware

End-of-line tests take too long

What happens. Each unit goes through a test at the end of the line: sweeping speeds, loads and temperatures, waiting for steady states. The test takes minutes, and the line cycle follows the test.

What it costs. Throughput, test benches multiplied to keep up, and energy spent on the test itself.

Why the usual fix fails. Cutting test points risks missing a fault. Waiting less for the steady state gives noisy values. More benches are the expensive way around the problem.

How a fixed-price remote diagnosis checks it. From test records of many units I look at which test points carry information, and how well the final values can be estimated from the early part of the transient. The fixed-price remote diagnosis tells you whether a shorter test with estimation would find the same faults.

Suspension has to trade comfort against handling

What happens. A suspension that is soft is comfortable and rolls in corners; one that is stiff handles well and shakes the passengers. Adjustable dampers allow a choice, but the setting is made by mode buttons or fixed rules.

What it costs. Customer complaints about either harshness or wallowing, and development effort on tuning every mode.

Why the usual fix fails. Rule-based switching reacts to what has already happened. More modes make the tuning harder. A sensor for the road ahead adds cost and still needs a control that uses it.

How a fixed-price remote diagnosis checks it. From vehicle test data (accelerations, damper travel, speed, steering) and the suspension model I look at how much comfort and handling could improve with predictive damper settings. The fixed-price remote diagnosis tells you what sensors and computing the control needs.

Chassis control is developed model-based but confirmed in road tests

What happens. Teams develop chassis control with models and simulation, but the last word comes from road tests. Every difference between the model and the car turns up at the test track, and engineers who can close that loop are scarce.

What it costs. Test days spent on parameter corrections, a development path that stretches, and open positions for control engineers that stay open.

Why the usual fix fails. More testing is slow and depends on weather. A better simulation does not help if its parameters are never matched to data. Hiring takes more time than the project has.

How a fixed-price remote diagnosis checks it. From test recordings and the chassis model I look at how well model and car agree, and which control parameters could be found from the model rather than on the road. The fixed-price remote diagnosis tells you how much road testing a model-based process could save.

Range-extender control needs people who know energy management

What happens. A range extender is a small engine with a generator that charges the battery of an electric vehicle. Deciding when it runs, at what power and for how long is an energy-management problem with noise, fuel and battery wear all counting. Teams with this experience are small.

What it costs. A strategy that runs the engine too often or at poor points, fuel and noise the driver notices, and a project waiting for people.

Why the usual fix fails. A simple thermostat rule (on at low charge, off at high) wastes the freedom the system has. Calibrating a map by road tests takes long and does not cover all routes.

How a fixed-price remote diagnosis checks it. From drive cycle data, engine and battery maps I look at what an economic operating strategy could gain over the current rule. The fixed-price remote diagnosis tells you whether such a strategy is worth implementing and what it needs.

Background: Economic predictive control

Control functions for combustion, hybrid and electric powertrains pile up

What happens. One supplier works on combustion engines, hybrids and electric drives at once. Each has its own control functions, often written separately, with the calibration done again for each variant.

What it costs. Engineering hours that multiply with every variant, inconsistencies between versions, and an effort to keep experts for all three worlds.

Why the usual fix fails. Copying functions from one powertrain to another carries assumptions that do not hold. Larger calibration teams slow the project down. Hand-tuned tables hide why a value was chosen.

How a fixed-price remote diagnosis checks it. From the function descriptions, parameter sets and test results I look at which parts are common and which depend on the powertrain, and where a model can replace tables. The fixed-price remote diagnosis tells you whether a model-based structure would cut the effort.

Background: Calibration-efficient predictive control

Safety-critical embedded control software needs specialists at short notice

What happens. A project has safety-critical control software for an embedded controller and needs experienced people quickly, often on a contract for a limited time. The knowledge is spread over few people, and the schedule is fixed.

What it costs. Delays while searching, schedule risk, and a team that cannot absorb a newcomer at full speed.

Why the usual fix fails. Agencies deliver applications, not understanding of your function. A new person needs weeks to read the existing code. Splitting the work without clear interfaces creates integration problems.

How a fixed-price remote diagnosis checks it. From the requirements, the architecture description and sample code I look at how the control function can be split into pieces that can be built and tested separately. The fixed-price remote diagnosis tells you what could be handed over with a clear scope, and what stays with your team.

Background: Predictive control on embedded hardware

Diversifying beyond cars brings products with unfamiliar dynamics

What happens. Many suppliers look beyond cars for new business: energy, aviation, rail or defence. Their strengths are manufacturing and cost, but the new products have different dynamics, control tasks and certification demands.

What it costs. Development risk on products the company has not built before, slow certification work, and no control engineers who know the new field.

Why the usual fix fails. Hiring a specialist team is slow and expensive for a business that has not yet won an order. Adapting a car controller to another field takes it outside the conditions it was proven in.

How a fixed-price remote diagnosis checks it. From the product requirements, the plant data or its model and what exists of the control I look at what the new product needs to hold: limits, accuracy, safety. The fixed-price remote diagnosis tells you what control and estimation the new product class needs, and what you can build with your own people.