Table of Contents

Machining Issues Outside the Model

Some machining problems the simulation does not reproduce, or reproduces only in part. Each section below says what HiNC reports near the problem and what it leaves out. Where a published, openly licensed measurement shows how large the unmodelled effect can be, the section quotes it. Where an effect is left out, read a result as a statement about the cutter, the spindle and the programmed path, not about the part or the machine that cut it.

Ordered from the part outward: the part's own stiffness and how its modes and its support change, the stress it carries, the finished surface, the cutter's life and the coolant, then the spindle and the machine structure. The last section is how to hand an intermediate workpiece to an analysis that models what is left out here. None of the cases cited here was simulated for this page; every number quoted from them is the publisher's measurement, unless the section says it is the authors' own calculation, and the sources are listed at the end.

Thin Walls and Workpiece Deflection

The workpiece in the simulation is rigid: it does not bend away from the cutter, and its vibration is not estimated, as Chatter, and What the Simulation Does About It states for workpiece chatter. On a thin wall the removed material is therefore the programmed engagement, not the smaller engagement a deflecting wall leaves.

A published test shows the size of the difference. A Ti-6Al-4V wall 125 mm long and 60 mm high was finished from 2.6 mm to a 2.2 mm target in two 30 mm axial levels with a Ø16 end mill, down milling at a 0.4 mm width of cut. Cut conventionally, the wall ended 330 µm too thick at the top at mid-length and 360 µm too thick near its end; repeating the path twice without further infeed brought the first to 270 µm. A control loop that measured the wall during the cut and added engagement brought the two to 80 µm and 130 µm [1].

Read the simulated force on such a part as the load on the cutter and on the wall. How far the wall yields under it has to be judged outside the simulation.

Workpiece Modes as Metal Is Removed

A rigid workpiece has no natural frequency, so the simulation cannot report one, and the frequency does not move as metal is removed. On a thin part it does move, and not always in one direction.

Hammer tests on a TC4 (Ti-6Al-4V) thin-walled workpiece, clamped in a vice and milled from both sides with a Ø10 four-flute end mill at 6,000 rpm, 500 mm/min and a 0.2 mm width of cut, were made at six stages: before any cutting, and after the cut had advanced 10, 20, 30, 40 and 50 mm along the workpiece. The measured first natural frequency (Table 1, column ωs) was 725.2 Hz before cutting, then 777.8, 787.5, 781.4, 710.1 and 627.9 Hz. It rose, and then fell. The authors' finite-element column for the same six states runs 716.0, 761.8, 781.5, 754.9, 676.7 and 573.6 Hz; that column is their calculation, not the hammer test [2].

The same movement is why a 2 mm floor has to be measured more than once. Square plates of UNS 2024-T3 aluminium, 85 mm by 85 mm and 2 mm thick, were screwed down at the four corners only and a 50 mm pocket was milled dry with a Ø10 bull-nose end mill at 4,000 rpm. The frequency response was measured before machining, after the 5th pass, after the 17th, and with the pocket finished, because the removal changes the modes. The text states two frequencies: the first natural frequency of the plates was around 1,730 Hz before machining, and a second was 4,000 Hz [3].

Distortion After Unclamping

The simulation removes material geometrically. The part carries no stress field, so neither the stress the cut leaves in the surface layer nor the distortion that appears when the part is unclamped is estimated.

Hole-drilling measurements on 7050-T7451 aluminium blocks milled with four combinations of cutting speed and feed per tooth, with the same Ø12 end mill, put the machining-induced residual stress peak at 115–185 MPa and its penetration depth at 56–180 µm, depending on the combination. Wafers 1 mm thick cut from the milled blocks distorted, less where the measured stress layer was shallower [4].

Clamping and Fillers

The fixture in the simulation is a solid on the machine chain. It places the workpiece, and it can be collided with. It carries no clamping force, no stiffness of its own and no filler in the cells of a part.

Filling those cells changes whether a honeycomb survives the cut. Pieces of 7075 aluminium honeycomb, 190 mm by 50 mm by 15 mm, were milled while the cells were full of ice — water frozen over the piece and kept frozen by liquid nitrogen — with a Ø16 cutter carrying ten inserts. Every measured cut in the series was filled that way. Across the sixteen cuts the roughness average Ra ran from 0.288 µm to 0.612 µm (Table 4); a confirmation cut with another cutter, at the parameters the series favoured, measured 0.218 µm. Unsuitable parameters still left cracks and burrs on the ice-filled surface (Figure 6). The authors' own cutting simulation, which is not a measurement, shows the unfilled cell wall bending and a serious defect on the machined surface, and the ice-filled wall without that defect (Figure 2) [5]. The simulation here cuts the honeycomb geometry with nothing in the cells, so it cannot tell that surface from the filled one.

Surface Roughness Values

The Surface Roughness chart (see Strip Charts) plots the geometric and deflection contributions to the finished surface in micrometres: the re-cut depth, the program-side cusp and the change in tool tip deflection. It does not compute a roughness parameter such as Ra or Rz, and it has no series for the cusp left between neighbouring passes.

Ball-end finishing of convex and concave R45 faces in 42CrMo4, in five raster directions at a 0.15 mm stepover, gave an average Rz that rose from about 3 µm to about 8 µm on the convex face as the raster turned from along the cylinder axis to across it, while the concave face stayed between about 4 µm and 6.3 µm (read from the published chart) [6]. The chart's trend does not follow such a measurement either: rebuilt in HiNC 3.2.42, the change of tip deflection the chart plots fell as the measured Rz rose, while the variation of the cutting force along each pass, which the chart does not plot, rose with it (a rank comparison over five directions; see Ball-End Finishing in Five Directions).

Tool Life and Fatigue

The simulation reports the load that risks an instantaneous failure and the heat that shortens tool life — the stress, spindle torque, spindle power and thermal stress ratios of Evaluating Process Machinability, on which a value over 100 % is bad machining, not necessarily failed machining (Bad Machining, Not Necessarily Failed Machining) — and wear quantities, described in Tool Life and Wear. It does not count the cycles a cutter survives, and it has no fatigue or cumulative-damage model. Nor does it count revolutions: a stress ratio at or above 100 % is bad machining, and how long a cutter holds there is not modelled. The holder, the spindle and the machine are rigid in the model: their compliance is not in it.

A run-to-failure data set in 42CrMo4 hardened to 38 HRC, with Ø10 end mills on one contour at 3,200 rpm and 640 mm/min, shows the spread such a count has to cover. At an 8 mm radial depth of cut the tools lasted 29–53 cycles before failing; at 4.5 mm, 59–149. With the holder length raised from 80 mm to 160 mm at 4.5 mm, one tool lasted 14 cycles and the other failed in its first [7]. The ratios can rank such configurations by load; the number of cycles is outside the model.

Coolant, Lubrication and Cryogenic Cooling

The coolant M-codes select a convection coefficient and nothing else. M08 uses the flood baseline of the chosen coolant, M07 that baseline times the mist ratio, and M09 the coefficient for coolant off, as Coolant Model sets out. The choice is not a friction coefficient, not a lubricant, and not a cryogenic temperature at the edge.

Milling of AISI 1045 at 220 HB, with a Ø16 cutter and two coated carbide inserts, compared minimum-quantity lubrication — a canola oil at 100 mL/h — with the same oil plus carbon dioxide at an injection pressure of 10 bar, which they state was −78 °C once it reached the tool tip. Over a cut 7,245 mm long, with the cutting speed raised during the test from 340 m/min to 400 m/min, the wear at the end was 27 % higher under the oil alone than under the oil plus carbon dioxide; earlier in the cut the difference was at most about 20 % (Figure 2). The root-mean-square cutting force stayed within 10 % between the two (Table 3) [8]. A convection coefficient can move a temperature. It does not apply a lubricant, and it does not put the edge at −78 °C.

Spindle Power at Idle

The spindle input power the simulation reports is the cutting power divided by the spindle's energy efficiency, as Spindle Capability derives it, and it leaves out the dry-run power that bearing friction and windage take while the spindle turns; so do the power ratios. The spindle meter power adds the dry-run power, and it is the value to compare with a power reading on the machine. It comes only as close as the spindle's two dry-run coefficients are to that spindle. The shipped defaults are rough estimates from published studies of bearing friction and gear windage, not measurements of a spindle, and give 612 W at 18,000 rpm. Fit them to the machine's own readings in air, as Fitting the dry-run coefficients shows, before comparing.

A 1983 high-speed machining trial on an 18,000 rpm spindle rated 16.6 hp shows how large the idle share can be. The roughing cutter drew 12.0–13.5 hp on a 0.300 in deep pass at 150 in/min, and still 7.1–7.2 hp on a 0.054 in deep pass at 200 in/min; the report does not separate out the idle power [9].

Machine Geometric Error

A constant geometric error of the machine — the squareness of two linear axes, the tilt of an axis, a rotary axis' pivot offset, an offset of machine zero, the tilt of the spindle housing — can be put on the machine chain as an error-term branch (see Error-Term Branches for the model and Error terms for the steps). The controller solves machine coordinates on the nominal chain — the same chain with every error-term branch taken as No Transform — while the simulated machine moves with the errors. The errors are therefore cut into the simulated part, as they are on a real machine whose controller does not know them, with or without RTCP and tilted work planes, and in CL playback on a machine tool alike. With the project's error terms switched off, the machine runs without them as well, and the part carries none. That run's cut, the nominal cut, is the reference a geometry difference reads the errors' effect against. The values have to come from a measurement of the machine; HiNC does not estimate them.

One difference from a real machine lies in the work offset: a touch-off there finds the part where it really is, so its offset holds the shift the machine's errors cause with every axis at zero, whereas HiNC measures program zero on the nominal chain, so that shift stays in the simulated cut. To match a given shop's part, type the work offset that shop measured; see Program Zero Alignment.

Errors that vary with position — straightness, pitch and scale errors, a volumetric error map — are not represented. Neither are reversal and backlash, servo mismatch, or the compensation tables of a real controller; thermal growth has its own section below. How a worn machine enters the optimizer instead is in Machine Condition and Safety Factors, and what that still cannot reproduce is the next section.

A machine's volumetric error combines, at each position, the position-dependent errors left out here and constant ones, such as squareness, that an error-term branch can carry. On a three-axis machining centre, compensating the volumetric error measured with a laser tracer took it from 49 µm to 14 µm. On a part finished by circular milling before and after the compensation, the mean roundness over seven measured sections improved from 10.13 µm to 6.16 µm on its Ø300 mm diameter and from 9.39 µm to 5.71 µm on its Ø200 mm diameter, while its Ø100 mm diameter stayed at about 5.4–5.6 µm [10].

Faults a Safety Factor Cannot Express

An error term carries a constant. Wear, and the rest of a machine's condition, stay with the safety factors: raising one lowers the feed the optimizer will assign, across the whole program, so the program leaves the worn machine more headroom (Machine Condition and Safety Factors). The simulated part does not change with them. The two holes are still where the nominal chain put them.

A fault that grows with wear, and that repeats with the angle of a rotary axis, is neither a constant nor a feed. Two holes of Ø200 mm, one milled, the rotary table turned 180°, and the other milled, were required to be coaxial within Φ0.01 mm. Twenty-four readings of that coaxiality had a mean of 12.54 µm and ran from 5 µm to 20 µm, so the mean was already past the allowance and the largest reading was twice it. Clamping, earlier operations and the cutting conditions were ruled out — a dial indicator on the clamped wall stayed within 0.01 mm, and easing the cut did not remove the error. The vibration of the table then showed an impulse once per revolution of its worm, at the worm's rotation frequency of 13.28 Hz, which the authors read as severe wear of the worm gear and the clearance it leaves in the mesh, and as the cause of the missed coaxiality [11]. A larger safety factor can ask for more margin. It does not put the second hole out of line with the first.

Machine Thermal Drift

The temperatures the simulation computes are those of the cutting zone — cutter, chip and workpiece surface — and the spindle body's thermal envelope; see Temperature and Wear. The thermal growth of the machine structure is not modelled, so the tool-to-part position does not drift over a run.

Capacitive sensors at four targets near the corners of a machine table measured how far the spindle moved relative to the table while the X axis warmed up with repeated 24 in rapid moves: about −160 µm in X at the two targets on one side of the table and about −50 to −60 µm at the other two, with Y and Z near zero [12].

Handing an Intermediate Workpiece to an External Analysis

HiNC does not estimate the compliance of the workpiece or the stress a cut leaves in it, so neither is a result that can be handed to another program. What can be handed across is the workpiece geometry at a chosen point in the mission, and the force on the workpiece through the cut.

Export Meshed Geometry (STL) and Record Meshed Geometry are commands in the mission list, in the Output category, and each one runs where it stands. Put an export where the stock is the one the other analysis should start from — after a roughing program, and before the next. The path is relative to the project folder and defaults to Output/MeshedGeom.stl; give each stage its own name, such as Output/after-rough.stl, so a later export does not replace the file. The resolution, in millimetres, is how finely that mesh is tessellated; 0 uses the display mesh resolution. The file is a binary STL of the workpiece mesh as it stands when the entry runs. Record Meshed Geometry writes that same mesh in HiNC's own format, Cache/Workpiece.wct by default, for a later run of HiNC to read back. If the tool that meshes the part for the other analysis prefers triangles whose size changes gradually, tick Even Mesh on the export: the STL holds somewhat more triangles, and the workpiece is not changed. The paths, the four record actions, what the resolution does, and what Even Mesh changes and does not, are in Recording a Play and Keeping the Workpiece.

The force through the cut is a shot file, and it is not one of those two commands. Under Post-Execution, switch on Shot Files Output. It is listed only while Show Physics Options is on, and the workpiece columns are the cutting force only on a play that ran with the physics on; without that they are zero. The template defaults to Output/[NcName].shot.csv. [NcName] is the program's file name with its extension kept, so NC/circle.ptp is written as Output/circle.ptp.shot.csv. One file is written per program, and it is opened afresh whenever the steps turn to that program. The sampling period is in milliseconds and defaults to 1. A row is FileNo, LineNo and Time — the simulated time, as a time span — then one column for each axis of the machine chain, in the chain's own order (X, Y, Z, A, B, C on a general XYZABC chain), then Tool.Fx, Tool.Fy and Tool.Fz, the force on the tool in the tool's running coordinate, Workpiece.Fx, Workpiece.Fy and Workpiece.Fz, the force on the workpiece in the program coordinate, in newtons, and Holder.Mx, Holder.My and Holder.Mz, the moment about the tool's observation point in the spindle's rotation coordinate, in newton-metres. A play of an NC program adds no ActualDateTime column. Workpiece.Fx, Workpiece.Fy and Workpiece.Fz, read against Time, are the load history. The entry has to sit below the program it covers: it writes the steps already played. The period, the file size and a write that finds nothing to write are in The Other Commands.

Sources

The numbers above are quoted unchanged from these sources, or read from their charts where the section says so. Each is provided by its authors or publisher as is, without warranty, and none of them endorses HiNC.

  1. L. Evers, M. Müller, C. Möller, A. Brouschkin, J. H. Dege, “Control Loop Based Dimensional Error Compensation for Milling of Near-Net-Shaped, Thin-Walled Structures”, in H. Kohl et al. (eds.), GCSM 2024, Lecture Notes in Mechanical Engineering, Springer, 2025, pp. 399–407, https://doi.org/10.1007/978-3-031-93891-7_44 (open-access copy: https://doi.org/10.15480/882.15340). © The Author(s) 2025, licensed CC BY 4.0. Values from Section 4.1.
  2. Y. Li, F. Ding, W. Tian, D. Wang, J. Zhou, “Prediction of modal parameters for thin-walled blade milling process considering material removal effect”, PLOS ONE, 9 September 2025, https://doi.org/10.1371/journal.pone.0323871. © 2025 Li et al., licensed CC BY 4.0. Measured frequencies from Table 1, column ωs; the finite-element column cited beside them is ωA.
  3. M. Casuso, A. Rubio-Mateos, F. Veiga, A. Lamikiz, “Influence of Axial Depth of Cut and Tool Position on Surface Quality and Chatter Appearance in Locally Supported Thin Floor Milling”, Materials 15(3), 731 (2022), https://doi.org/10.3390/ma15030731. © 2022 by the authors, licensee MDPI, licensed CC BY 4.0. The four measurement stages are in Section 2.4; the two stated frequencies are in the discussion of Figure 8.
  4. D. Weber, B. Kirsch, C. R. Chighizola, C. R. D'Elia, B. S. Linke, M. R. Hill, J. C. Aurich, “Analysis of machining-induced residual stresses of milled aluminum workpieces, their repeatability, and their resulting distortion”, The International Journal of Advanced Manufacturing Technology 115, 1089–1110 (2021), https://doi.org/10.1007/s00170-021-07171-7. © The Author(s) 2021, licensed CC BY 4.0. Values from Table 7.
  5. Y. Wang, Y. Gan, H. Liu, L. Han, J. Wang, K. Liu, “Surface Quality Improvement in Machining an Aluminum Honeycomb by Ice Fixation”, Chinese Journal of Mechanical Engineering 33, 20 (2020), https://doi.org/10.1186/s10033-020-00439-1. © The Author(s) 2020, licensed CC BY 4.0. Ra from Table 4 and Section 4.2; Figure 2 is the authors' simulation, and Figure 6 the morphology.
  6. B. Mikó, B. Varga, W. Zębala, “The Effect of the Feed Direction on the Micro- and Macro Accuracy of 3D Ball-end Milling of Chromium-Molybdenum Alloy Steel”, Materials 12(24), 4038 (2019), https://doi.org/10.3390/ma12244038. © 2019 by the authors, licensee MDPI, licensed CC BY 4.0. Values read from Figure 7.
  7. G. Piecuch, T. Żabinski, “A new open dataset from a milling process – data for classification and estimation of tool life”, figshare data set (2025), https://doi.org/10.6084/m9.figshare.28589216.v1, described in Scientific Data 12, 650 (2025), https://doi.org/10.1038/s41597-025-04923-y. © The Author(s) 2025, licensed CC BY 4.0. Values from the CycleToFailure column of the data set's metadata, counted as the cycles a tool ran before the failing one.
  8. N. Villarrazo, S. Caneda, O. Pereira, A. Rodríguez, L. N. López de Lacalle, “The Effects of Lubricooling Ecosustainable Techniques on Tool Wear in Carbon Steel Milling”, Materials 16(7), 2936 (2023), https://doi.org/10.3390/ma16072936. © 2023 by the authors, licensee MDPI, licensed CC BY 4.0. Wear from the discussion of Figure 2; the force comparison from the discussion of Table 3.
  9. J. D. Hankins, “High Speed Machining of Space Shuttle External Tank Liquid Hydrogen Barrel Panel”, NASA-TM-82557, NASA Marshall Space Flight Center, November 1983, https://ntrs.nasa.gov/citations/19840006284. Work of the U.S. Government, public use permitted (NASA STI terms). Values from the horsepower data sheets, Tables 3 and 4.
  10. M. Holub, R. Jankovych, J. Vetiska, J. Sramek, P. Blecha, J. Smolik, P. Heinrich, “Experimental Study of the Volumetric Error Effect on the Resulting Working Accuracy—Roundness”, Applied Sciences 10(18), 6233 (2020), https://doi.org/10.3390/app10186233. © 2020 by the authors, licensee MDPI, licensed CC BY 4.0. Values from Figure 7 and Tables 11–13.
  11. H. Cao, D. Li, Y. Yue, “Root Cause Identification of Machining Error Based on Statistical Process Control and Fault Diagnosis of Machine Tools”, Machines 5(3), 20 (2017), https://doi.org/10.3390/machines5030020. © 2017 by the authors, licensee MDPI, licensed CC BY 4.0. Coaxiality from Section 3.1; the worm-gear reading from Section 3.3.
  12. G. W. Vogl (2023), “Thermal Drift Monitoring Experiment 01”, National Institute of Standards and Technology, https://doi.org/10.18434/mds2-2934, a work of NIST employees, not subject to copyright in the United States (NIST data terms). The warm-up displacement is from Figure 8a of G. W. Vogl et al., “Vision-based thermal drift monitoring method for machine tools”, CIRP Annals 72 (2023) 301–304, https://doi.org/10.1016/j.cirp.2023.04.053.

See Also