However, during this time, it has also been found that HSM poses a new set of problems that can be very costly. These can lead to very rapid spindle failures and to scrapped parts due to chatter. This article describes two systems developed and in use by the Boeing Commercial Airplane Group a sensor based adaptive control system and the Machining Prediction Software (MPS) used to predict chatter and other process problems. Both have proven to be of crucial importance for reducing cost of both NC programming and operation of HSM.

Over a decade ago, the Boeing group began work on Machining Prediction Software (MPS). In particular, they researched regenerative chatter in cuts with large axial depths and small radial depth, which are commonly used in airplane part manufacturing. A new type of chatter called avalanche chatter was also analyzed and prediction software for that was developed.

This very early version of MPS was important for the evaluation of the adaptive control (AC) in the shop. It allowed optimal cutter force limits to be calculated and correct speed to be selected. It should be noted that at that time HSM was not yet implemented in Boeing’s shops in Auburn, and the prediction software was applied in the lower speed range used then on many machines.

Applications of MPS In Production at Boeing

MPS was first used in production NC programming at Boeing in August 1992. It proved to correctly predict regenerative and avalanche chatter very common problems then and still today and was soon accepted. Training of NC programmers was started on a very small scale. But after some time, it was obvious that most programmers did not use MPS before they had run into chatter or other problems during the tape tryout (TTO). Due to the extreme pressure programmers felt in producing part programs quickly, MPS was not used to deliberately optimize selection of cutters and cross sections before writing the program. Programmers preferred to go through several cycles of TTO adjustments than to run MPS when the programs were not 100% optimal.

In the wing manufacturing shop, there was another problem. A new machining cell with two large three-spindle machines had been installed early in 1993. The spindles have 10,000-RPM range and 100 Hp capacity. As of June that year, no parts had been produced due to chatter problems. Attempts to cut the parts had instead destroyed six spindles over a half year period. These poor results showed a drawback of HSM most programmers and machinists can’t avoid chatter without using analytical software like MPS. After that first experience with the destroyed spindles, the programmers asked for MPS to be used in their cell and very quickly made good parts. In spite of that, it was very soon realized that the spindle tapers wore out early with 20-30 hours of machining necessitating disassembly and re-grind, a very costly operation.

This fact demonstrated another drawback of HSM that in many cases, one is forced to accept a compromise between material removal rate and spindle life. This is because the forced vibrations from the cutting edges hitting the parts excites the cutter-holder structure so much that the allowable bending moment at the gage line is exceeded. This limit, based on experience data, is as low as 6,000 in.-lbs. for a 50 taper interface.

This caused a limited removal rate on many cuts in the wing shop. Instead of the 100 Hp available, only 10-20 Hp (sometimes up to the 40 Hp range) could be applied to the cutter. The corresponding removal rate was much lower than what had been expected before the reality of HSM was understood.

What was needed was software that could also predict bending movement at the gage line and also the load on the front bearing (L10), since that limit can also easily be exceeded. Amplification of forces due to structure resonance also had to be taken into account.

MPS was applied to many other HSM cells at Boeing and also to machines with lower speeds. In most cases, it had been used only as a trouble-shooting tool and not for planning the machining concepts as it was intended to do. A large effort to train Boeing’s NC programmers and shop personnel in the use of MPS was soon undertaken.

MPS has been used to design new cutters with inserts, which are used for spar chord machining. These cutters replaced large ones with brazed-on carbide edges that were made in-house. Cutter structures were made by the vendor, first without the costly cutting of flutes and seats to hold the inserts. A modal analysis of the blanks when installed in the spindle was made, and MPS was used to determine how many flutes were to be cut to fit the low speed range of the 4,200 RPM available on these machines. The vendor could then manufacture the cutters complete. Chatter-free cutting resulted with this method.

Users also required chatter prediction capability for machining weak part structures. An example of this is spar chords when setup so they are clamped on one leg with a dovetail on the extrusion angle. Another example is for machining of so-called T-chords, used for attaching the lower wing surface the side of the body. These parts, about 4-5 feet long, are held only at each end. As the stock material is cut away, the part gets weaker in the middle span area, so chatter develops. Correct speed prediction resulted from using the MPS program with part dynamics data from tap test.

Input Data for MPS

Physical models require a set of input variables which are needed for calculation of output data. MPS input variables include: cutter holder structure, part structure, cutter dimensions and static deflection of thin part flanges.

Material files contain specific or normalized cutting force per unit length of chip in contact with the edge. Two force components, one in tangential direction normal to the cutter radius and one in the radial direction are listed in these files as a function of chip thickness and cutting speed. The force data are measured with a dynamometer during a few revolutions of a single flute cutter in a special slotting cut. Software processes the dynamometer data to the format needed in the files. One file is needed for each combination of a specific edge geometry and a part material alloy.

Data for the cutter holder structure is measured with so called "Modal Analysis" techniques. A small hammer with a force sensor is used to tap the cutter, hence the term "Tap Test" used for this type of measurement. The vibration signature is picked up by an accelerometer attached temporarily to the cutter.

Commercially available spectrum analyzers were first used for tap testing. They are difficult to use for shop technicians, unless they have some engineering experience. Special software was therefore developed to fit the special requirements for MPS. It runs on a notebook computer and uses a PCMCIA card for interface of the analog data from the force and accelerometer sensors. In addition to being of much lower cost, the user interface is designed so that an engineering background is not required to operate it.

The tap test is also used to measure the dynamic behavior of the part-fixture structure if that has been determined to cause a chatter problem. Often, the problem occurs only after some material has been removed from the part so it is less rigid. The tap test must then be done at that sequence in the part program.

Cutter dimensions include diameter, flute length, number of flutes, helix angle, set-length, shank stick-out and dimensions of inserts, if used. The dimensions and dynamic data are stored in files, with the same names but with different extensions, or in a database.

Static deflection of thin part flanges is calculated with FEA and put into a file format that MPS can use.

The Current Version of MPS

Examining Optimal Machining Parameters for a Given Cutter

Assume here that the cutter data files as mentioned above have been created and that a material file is available for the edge to be used.

In HSM, the regenerative chatter problems are of primary concern. A program module of MPS addresses that subject so that program is brought up first. After entering in "best guess" data on a setup menu, the user can select either an interactive mode or a spreadsheet output mode. In the first mode, the feed, speed, axial depth of cut (AD) and radial depth of cut (RD) can be changed interactively, while observing a Nyquist Plot which shows if chatter will occur or not.

The user can change the cutting parameters, axial depth, etc., by entering data or using up-down arrows on the fields above the chart. If the cut is not chattering, the plot must not encircle the point (-1,0) on the horizontal axis. The point is surrounded by small rectangle serving as a margin.

The speed and other parameters are adjusted in the figure 1 so the point (-1,0) is not encircled, the trace just touches the rectangle, which means chatter-free cut. A small change in speed results in the plot in figure 2. Notice how the trace is now encircling the point (-1,0) indicating chatter will occur.

It is in some cases important to know the predicted frequency of chatter when analyzing process problems. The parameter along the trace is the frequency in Hz for a possible case of vibration. In the last case, a cursor in the form of a cross (X) is positioned so it is very close to the rectangle. The predicted frequency of the chatter is shown in the window below the slider for X cursor control, in this case 538.1 Hz. It could be that the part dynamics also contribute to chatter, and in that case the chatter frequency will be different. Another MPS program is used for chatter prediction in that case. It requires a tap test to be done on the part itself, as was briefly mentioned earlier.

The chatter prediction theory is based on mathematical stability criteria of closed loop systems.

If HSM of aluminum is analyzed, then the program discussed above is used to find optimal speed to avoid chatter at a maximum cross section. Once that is found, one has the choice of going to program 4, which calculates forces, bending moment at gage-line, bearing loads and cutter deflection. This program can be started by clicking on the button "Run Program 4," in the lower right corner of the screen. All the parameters set up interactively on the screen in the figures shown are also used in program 4.

Program 4 output is shown in figure 3. It has three plots, the upper left showing calculated cutting forces in X (feed direction), Y direction, and resultant force in the X-Y plane versus cutter rotation. This represents the force as a function of time, since speed is constant. A small runout of .001" was entered in the setup which made the peak values for the three flutes slightly different.

The chart on the upper right shows the bending moment at the gage-line. That moment should be within a certain limit depending on what tool holder interface is used. For 50 taper type, 6000 in-lbs. is considered to be a suitable limit, as compared to 12,000 in-lb for HSK 100A.

As the force varies, the cutter is deflected both in the feed direction and normal to the feed direction, Y. The chart to the lower left shows the deflection of the cutter in the Y direction versus cutter rotation. The X-axis is scaled so the Y-coordinate represents the material left or undercut on the surface. If the trace value is positive, material is left, else undercutting has been done, that is, the surface is below the programmed, theoretical value. The left end of the trace is at the tip of the cutter, and the right end corresponds to the axial depth.

At high speeds, the cutter deflects often a substantial amount due to dynamic forces magnified by the resonance in the cutter-spindle structure. The dotted trace includes this dynamic deflection. A slight increase above the static deflection can be seen in this case. Please see lower left chart in figure 3.

Depths of cuts, feed and speed, can be changed interactively so the effect can be viewed in a split second. The number of flutes and helix can also be changed temporarily to aid selection of best cutters. Chatter problems after changes have been made can be examined by a click on "Return to Nyq.plot." Many times several iterations are needed to find optimal cutting parameters and cutters.

Program 4 can also be setup to make one roughing cut, followed by a finishing cut and then one or more spring passes. This is of importance for cuts with long axial depths, particularly in hard metals, steel, titanium, etc., where cutter deflection can cause inaccuracy. The material left due to cutter deflection on the preceding pass is taken into account when processing deflection for the current pass.

Process Parameter Variations

MPS can also be setup to examine the effect of variations in process parameters that are difficult to control. Runout is one such parameter. A value of 0.001 inch Total Indicated Runout (TIR) was setup in the cases discussed so far. Another variable is cutter wear. It is examined by selecting a material file for a worn edge. For HSM, it is also important to examine the effect of changes in resonance frequencies (mode frequencies) of cutter-spindle dynamics. They can change relative to the tap tested data because of wear on spindle-holder taper interface, bearing wear, and pre-load. Some of the modes depend on bearing stiffness, which can change with spindle speed and bearing load. The tap testing is done with the spindle not rotating wand without side load. All of these variations will cause premature wear of machines or scrapped parts due to poor quality of machining.

Multiple-Constraint Prediction Program

After the user has determined the optimal speed for a HSM cutter, it is often useful to run a different program, No. 6, which predicts all six process problems over a range of axial and radial depths, selected at setup.

After setup and launch, there are no more interactions with the user. The result is instead provided as a text file containing one or more spreadsheets. The speed and feed rate is constant for each spreadsheet, but several sheets with different speed and feed can be processed in one run. The spreadsheet has the radial depth of cut (RD) in horizontal direction (columns) and axial depth of cut (AD) in vertical direction (rows). Each cell has six places for a character that is assigned for each of the different process problem analyzed by MPS.

The user inputs the limits of bearing load, bending moment, cutter deflection and available power to the cutter. If a problem is found for the particular cut represented by the cell location, then the particular character for that problem is printed.

Prediction software alone though a very important tool to find optimal cutters and machining parameters is not enough to avoid process problems in a production floor environment. The ideal solution to avoid problems is to use both the prediction software and sensor based adaptive control.

Sensor Based Adaptive Control with Tool Wear Indication

Programming Considerations

When adaptive control and MPS are available, several of the NC-programmer’s concerns are removed. No longer is that necessary to try to guess if an overload will occur, nor is it necessary to make several TTO’s to find the best feed, speed, and cut cross-sections to avoid chatter. The result is a large improvement in the productivity of the programmer and the shop.

The adaptive control makes cuts of changing cross-section much more effective because of the automatic feed rate adjustment. This means improved removal rate, and also longer tool life, because the number of revolutions with the flank rubbing during light cuts is reduced by a large factor two to four, typically. The programmer is not as constrained in selecting cutting concepts with adaptive control.

The availability of MPS also allows selection of a much larger cut cross-section without the risk of process problems during TTO. Programming for HSM of aluminum has been dominated by a doctrine of small axial depths typically 0.1 inch, and full radial depth, or a shallow slotting cut. The reason for this is that chatter is not likely to occur with such a small axial depth, at least not for stiff solid carbide cutters, such as one-inch diameter and three-inch length. If a two-flute cutter is used, the deflection of the cutter will be almost zero when the edge touches the cut surface. This means very small ridges between adjacent layers of cut in the axial direction.

Large cut cross sections reduce the total travel required to cut a part, reduce the feedrate and acceleration and, therefore, reduce wear and tear on the servo systems and the energy needed tout a part. This is of more concern in HSM of aluminum, where high feed rate and acceleration has been attempted lately. The engineering task of designing structures that do not flex and oscillate after sudden impacts of forces from 1g to 2g is very challenging, if not impossible.

Aluminum Pocket Cut
At the IMTS Show in 1998, Boeing used a Makino MC1013 machine to make an aluminum pocket cut programmed with MPS to demonstrate the function of adaptive control. Four pockets on each side of a horizontal 24" long aluminum bar were cut in one setup. This employed fairly long set-lengths typical when machining airplane structures, allowing access to many sides without tear-down and setup. A long set-length, in turn, can lead to large bending moments if adaptive control is not used, or else if extreme care and MPS are not applied.

All cuts were programmed with MPS. The cutter was selected to demonstrate a typical case where a programmer would have unknowingly overloaded the HSK100A interface, had not MPS nor the adaptive control been available. The heavy cuts were constrained by the bending moment, so override had to go down as far as 30 percent to avoid overload. A better cutter would have been one with 37.5 degree helix rather than 30 degrees that was used. The bending moment would have been reduced to only 25 percent at this very special radial depth, and a much higher removal rate would have resulted. The reason for less bending moment at 37.5 degrees helix is that the forced vibrations happen to be very small for that particular axial depth. That cutter would have made more justice to the powerful, 65 Hp spindle on the Makino MC1013 machine, but would have not demonstrated the function of adaptive control as well.

The cutting of corners as in this pocket has always given programmers problems (see Figure 4). The rapid buildup of radial depth due to the wrap-around effect often leads to chatter when large axial depths are used and very careful calculation of the path is not done. Several 90-degree arcs are often required to keep radial depth below the limit for chatter. In this case, five and seven arc moves in each corner were required to keep radial depth below the limit for chatter. A special program of MPS calculates location of the start points and arc radii for these moves.

Demonstration of Tool Wear Indication
Many process variables change over time in unpredictable ways, which makes programming of feed rates for every cross section of cut less effective as means of optimizing the process. If worst-case combination of variables would be applied, the removal rate would be unacceptably low, so some risk that overload will sometimes occur must be taken. One variable is the cutter flank wear. Every machine operator is well aware of this but still have little knowledge about the actual value at any given time. It can cause very large changes in cutter force, that is, in bending moment without any indication on the machine power meter or in form of noise or vibration. Adaptive control helps prevent severe overload during large changes in override and toolbar.

Commercial Version of Adaptive Control
The version of adaptive control demonstrated at IMTS ’98 used an interface to the CNC which does not fully take advantage of new, open architecture controls. For example, the serial port and analog I/O modules are used to transfer data between the AC system and the CNC. A new version uses a DSP plug in board in the PC. Instead of a serial port, the interface with the CNC will be over a high speed serial bus (HSSB) in case of Fanuc CNC’s or corresponding interface for other brands of CNC.

The advantage is lower cost, easier installation, less hardware and better reliability. It will still have Ethernet LAN link to the remote PC’s but the operator’s display which is running on the front end PC of the CNC will be interfaced directly to the DSP board via the PCI bus of the PC.

Conclusion
It has been shown how MPS is used to predict chatter and other machining problems. It is, therefore, and effective aid when selecting optimal cutting concepts, cutters, feed, speed, etc. However, it has also been shown again by using MPS that prediction alone is not sufficient for reliable machining operations because of uncontrollable process parameter variations.

The solution to this problem is to use both MPS for optimum programming and adaptive control to provide real time, sensor-based control of the process. This keeps the machine running at its maximum capacity without risk of destroying bearings, tapers and many other parts. Very large savings in machining cost are possible by using these two systems.

And that money saved leads to better profits.

Copyright © 2001 - 2008 by Makino, Inc. All rights reserved.