Machine tools: thermal drift, chatter, slow feeds

Spindle and frame heat up and the dimensions drift

What happens. After the machine starts, the spindle and frame warm up unevenly and the tool position moves by several hundredths of a millimetre against the cold state. Parts at the start of a shift differ from those later.

What it costs. Scrap and rework on tight tolerances, and a warm-up time before the first good part. Customers add their own warm-up routines.

Why the usual fix fails. Cooling the machine bed and the spindle reduces the problem but does not remove it. Measuring a probe artefact at intervals corrects after the drift has happened.

How a fixed-price remote diagnosis checks it. From machine logs (temperatures, spindle speeds, axis positions) and a few measured deviations I fit a thermal model that estimates the drift and check how well a compensation would work. The fixed-price remote diagnosis shows how much of the drift can be removed in the control and which temperature sensors the model needs.

Background: Computing unmeasured states for control

Chatter appears at higher speeds

What happens. At higher spindle speed or depth of cut the tool starts to vibrate, the surface shows waves, and the noise rises. The operator lowers the speed until it stops.

What it costs. Surface quality, tool life and machine wear, and speeds below what the spindle could deliver.

Why the usual fix fails. Stability charts from tests give safe speeds for tool and workpiece combinations that were measured. Every new tool or clamping needs a new test.

How a fixed-price remote diagnosis checks it. From vibration and spindle-power records of real cuts and a few test cuts I identify the dynamics that cause chatter. The fixed-price remote diagnosis tells you whether a model-based damping or speed choice could shift the limit, and what it would need from the drive.

Background: Control that adapts to drift and wear

Press brakes: springback varies with the material batch

What happens. A sheet bent to the programmed angle springs back, and the amount depends on the material batch, its thickness and rolling direction. The first parts of a batch are out of angle.

What it costs. Angle errors, rejects and trial bends at the start of every batch. A skilled operator is needed to correct by eye.

Why the usual fix fails. Material tables give an average springback and do not know the actual batch. A measuring system on the machine corrects after the bend.

How a fixed-price remote diagnosis checks it. From bending records (programmed and measured angles, material data, force, stroke) I check how well springback can be computed and adapted from the first bends. The fixed-price remote diagnosis shows whether the machine could correct itself from the first part.

Background: Control that adapts to drift and wear

Large machine tools: heavier axes, lower natural frequencies

What happens. As machine tools get bigger, the moving masses grow and the machine structure becomes softer. The natural frequencies sink into the range of the control loops.

What it costs. Precision and speed: the controller must be slowed so as not to excite the structure, and path accuracy suffers at the working point.

Why the usual fix fails. Tuning the standard cascade with notch filters works for one position of the axes. A stiffer structure costs weight, material and money.

How a fixed-price remote diagnosis checks it. From drive data, frequency-response measurements and the structural model I check how a model-based feedforward and active damping would change the achievable path accuracy. The fixed-price remote diagnosis tells you whether the next machine size can be controlled without redesigning the structure.

Background: Precision control for robots and machine tools

Presses with higher tonnage and larger beds deflect and tilt

What happens. A press with a larger bed and higher force deflects in the frame, and the ram tilts under off-centre loads. The die sees a different parallelism at every stroke.

What it costs. Part quality, uneven wear of the die and set-up time for adjusting wedges and shims.

Why the usual fix fails. A stiffer frame is heavy and expensive. Corrections by hand at set-up apply to one load and one die.

How a fixed-price remote diagnosis checks it. From force, position and tilt measurements of the real press I compute the deflection and tilt as a function of load, and check whether the control can compensate them with the existing actuators. The fixed-price remote diagnosis shows how far the compensation can go and what sensors it needs.

Background: Precision control for robots and machine tools

Accuracy is lost at higher feed rates

What happens. At low feed the contour is accurate. As the feed rises, the tool cuts corners, circles become ovals and the path error grows.

What it costs. A choice between speed and precision. Customers buy a fast machine and then run it slowly for accurate parts.

Why the usual fix fails. Standard axis controllers follow each axis alone and do not know the contour. More feedforward helps only for what was tuned.

How a fixed-price remote diagnosis checks it. From the machine’s axis logs and contour measurements I check how much of the error comes from axis dynamics and how contouring control that regards the path as a whole would reduce it. The fixed-price remote diagnosis shows what accuracy at which feed rate is realistic for your machine.

Background: Contour accuracy against speed

Feed rates are set conservatively

What happens. The programmer sets feed rates that are safe for the whole part, with limits for the tightest corner or the thinnest wall. The machine could move faster in long sections.

What it costs. Cycle time on every part, summed over a year of production.

Why the usual fix fails. Raising the feed globally causes errors at the critical points. Setting it by hand for each section is slow and relies on trial and error.

How a fixed-price remote diagnosis checks it. From axis limits, drive data and the part programme I compute how fast the tool can go along each section while keeping the error in tolerance. The fixed-price remote diagnosis tells you how much cycle time a feed optimisation along the path would give.

Background: Optimal paths and setpoint plans

Thin parts deflect under the cutting force

What happens. Thin walls and slender parts bend away from the tool. After the cut they spring back, and the finished wall is thicker than programmed.

What it costs. Out-of-tolerance parts, extra finishing passes and slow cuts.

Why the usual fix fails. Extra passes cost time. Fixtures that support the part cost set-up work and do not suit every geometry.

How a fixed-price remote diagnosis checks it. From part geometry, cutting forces and a few test cuts I compute the deflection and check whether the tool path can be corrected in advance. The fixed-price remote diagnosis shows whether predicted and compensated deflection brings the part into tolerance on your machine.

Background: Precision control for robots and machine tools

Robots are too soft for machining

What happens. An industrial robot is accurate in repeating a position, but under cutting force its arm bends, and the tool leaves its path. Milling with a robot is cheaper than a machine tool, but the results are poor.

What it costs. Accuracy of the part, and surface quality. The robot is used for rough work only.

Why the usual fix fails. Slowing the cut reduces the force but adds time. A stiffer robot is more expensive and not available for every size.

How a fixed-price remote diagnosis checks it. From robot data, force measurements and the stiffness of the arm I build a stiffness model and check how far compensating the tool path improves accuracy. The fixed-price remote diagnosis shows which accuracy is realistic for milling with your robot.

Background: Precision control for robots and machine tools

Every machine is tuned by a service engineer

What happens. At commissioning a service engineer adjusts the axes one after another, runs tests and changes the gains. It takes days and results depend on who does it.

What it costs. Commissioning time per machine, travel and the dependency on a few experienced people.

Why the usual fix fails. A standard auto-tune function gives conservative results, and a tuning checklist gives a start, not an optimum. The more variants, the more people are needed.

How a fixed-price remote diagnosis checks it. From the axis response data that a commissioning run records I identify the dynamics of each axis and compute controller settings. The fixed-price remote diagnosis shows whether tuning can be automated from identification for your axes and how close it comes to what the best engineer achieves.

Background: Replacing hand-tuned parameter tables by a model

Machines idle at full readiness

What happens. Between jobs, at breaks and at weekends machines stay switched on, with pumps, fans, hydraulics and cooling running, so that they are ready at once.

What it costs. Power for no output. Across a hall of machines and a year, it is a considerable share of the energy bill.

Why the usual fix fails. Switching everything off by hand risks long warm-up times and thermal drift. A simple timer cannot know which machine is needed next.

How a fixed-price remote diagnosis checks it. From machine power logs and the job schedule I check which units can drop into standby and when they must wake up to be ready. The fixed-price remote diagnosis shows how much a standby strategy would save and what it needs from the machine control.

Background: Economic predictive control

Grinding: burn and dimension are judged by experience

What happens. The grinder’s wheel wears, the workpiece heats up, and the dimension changes during the process. Burn marks on the surface show up only in a check afterwards.

What it costs. Rejects, rework and a safety margin in the cycle. A burnt part may be found late, after further steps.

Why the usual fix fails. Fixed infeed and dressing intervals are based on experience and a margin. Measuring the part in the machine gives the size, not the temperature in the contact zone.

How a fixed-price remote diagnosis checks it. From power, force and acoustic signals of the grinding process and the inspection results I check whether the state of the process can be estimated during the cut. The fixed-price remote diagnosis shows whether a computed state can warn of burn and trim the dimension, and which signals are needed.

Background: Computed measurements

Laser cutting settings are found by trial for each material

What happens. For each new material, thickness and lot the cutting parameters are tried out on test pieces until the edge is clean. The table in the controller is a starting point.

What it costs. Set-up time and scrap sheets, and edge quality that depends on whoever tuned the parameters.

Why the usual fix fails. A parameter database collects results but covers only what was tried. A change of gas, nozzle or lens wear shifts the optimum again.

How a fixed-price remote diagnosis checks it. From the cutting logs (power, speed, gas pressure, results) I check whether a model of the process can propose settings for a new material and how many test cuts it would save. The fixed-price remote diagnosis tells you whether model-based settings are realistic for your cutters.

Spark erosion: the gap conditions are unstable

What happens. In spark erosion the gap between the electrode and the workpiece changes with debris, flushing and wear. The generator reacts to short circuits and arcs by pulling back.

What it costs. Speed: the machine runs below its capability to stay out of trouble, and unstable gap conditions leave a worse surface.

Why the usual fix fails. Fixed generator settings per task cannot follow the gap. Adding a new generator type does not change that.

How a fixed-price remote diagnosis checks it. From voltage and current records of the process, I check whether the gap state can be estimated continuously and whether a control that uses this estimate could keep the process steady at a higher rate. The fixed-price remote diagnosis shows what that needs from your generator and control.

Background: Computing unmeasured states for control

Probing cycles on the machine are slow

What happens. Measuring a part on the machine takes many touch points. Each one means a move, a touch and a retract, and the cycle time rises.

What it costs. Cycle time and machine hours spent on measuring instead of cutting. If measuring is skipped, errors are found late.

Why the usual fix fails. Fewer points save time but lose information about the form. A separate measuring machine moves the work to another place.

How a fixed-price remote diagnosis checks it. From measured parts and the machine’s error behaviour I check whether a model of the part and the machine can reduce the number of points while keeping the information that matters. The fixed-price remote diagnosis shows how many points could be saved on your parts.

Many machine variants, each parameterised by hand

What happens. A machine range comes in many sizes and options, and each variant needs its own set of control parameters, which engineers enter and test by hand.

What it costs. Engineering time, which grows with each variant, and mistakes in parameter tables that surface at the customer.

Why the usual fix fails. Copying the values of the nearest variant gives a start that has to be corrected. Large tables are hard to keep consistent.

How a fixed-price remote diagnosis checks it. From the parameter sets and measurements of existing variants I check whether a model can generate the parameters of a new variant from its data. The fixed-price remote diagnosis shows whether parameterisation from a model would replace the tables for your range.

Background: Replacing hand-tuned parameter tables by a model

Faults are found at the first real run

What happens. Control software for a new machine is first run on the real machine. Errors in axis coupling, safety logic or sequences show up when the first parts are made.

What it costs. Delays at the end of the project, when pressure is highest, and a risk of damage to the machine.

Why the usual fix fails. Testing on a prototype takes weeks of machine time. Simulation that only plays back signals does not show whether the real axes would follow.

How a fixed-price remote diagnosis checks it. From the machine design, drive data and control programme I check which parts need a dynamic model to make simulation worth the effort. The fixed-price remote diagnosis shows which faults a simulation before the build can really catch on your machine.

Control software is the bottleneck in machine building

What happens. A large share of machine builders say staff shortages cost them sales (Produktion). Development moves from mechanics towards software and automation, and control engineers are among the hardest roles to fill.

What it costs. Projects wait for control software, and delivery dates slip. Work that could be sold stays unsold.

Why the usual fix fails. Posting another job and waiting for a hire takes months. A permanent hire for a peak in projects leaves capacity unused afterwards.

How a fixed-price remote diagnosis checks it. I take defined control problems on a fixed price, remotely, from your logs and drawings, or work with an agency arrangement for longer periods. The fixed-price remote diagnosis is the way to see quickly where your control software is limiting and what can be outsourced.

Rules for machines with learning functions are still being settled

What happens. Machine builders add learning functions to controls, from adaptive settings to image-based decisions. The European rules for artificial intelligence and for machinery overlap, and the deadlines for artificial intelligence in products have been postponed (Gibson Dunn).

What it costs. Uncertainty over which route a product with a learning function must take, and the risk of designing a function that has to be reworked.

Why the usual fix fails. Waiting for the rules to settle costs time to market. A learning function that cannot be explained or limited is hard to certify under any route.

How a fixed-price remote diagnosis checks it. I review which functions in your control could be built as model-based controllers with explicit limits instead of black boxes, from the specification and the data you have. The fixed-price remote diagnosis shows where that simpler route is enough.

Background: Control that adapts to drift and wear