Using Device-Specific Data Acquisition for Automated Laboratory Testing

Transcription

1 TRANSPOR'IATION RESEARCH RECORD Using Device-Specific Data Acqisition for Atomated Laboratory Testing THOMAS C. SHEAHAN, DON J. DEGROOT, AND JOHN T. GERMAINE Compter-based data acqisition systems are becoming common eqipment in the geotechnical laboratory. Althogh centralized, laboratoryide systems offer increased data accontability, flexibility, and access control, their high costs can be jstified only in large laboratories and they are not sited for se in feedback control of compter-atomated eqipment. A device-specific data acqisition system is dedicated to a single testing apparats and ths allos smaller laboratories to incorporate a data acqisition capability one device at a time. From a comparison of the to system types, it is conclded that only the devicespecific approach is appropriate for se in compter-atomated, feedback-based servo control of testing devices. Additional reqirements for device servo control are described, inclding servo control algorithms (proportional and proportional pls integral pls derivative) and other softare reqirements for control based on devicespecific data acqisition systems. Experiences in applying these basic principles to the compter atomation of a nmber of different geotechnical laboratory devices are described. Data from varios tests are provided to illstrate the atomated test control qality. Improved technology and loer costs for data acqisition hardare and compters have made device-specific systems an economical soltion for obtaining high-qality test data. Althogh compter-atomated geotechnical laboratory testing as a research topic 10 years ago, today the eqipment for sch testing is common in many research laboratories and in some commercial laboratories. Significant advances in and the redced costs of personal compter (PC)-based technologies are in part responsible for this rapid development and practical implementation. The other driving force behind compter-atomated testing eqipment is the desire to improve laboratory prodctivity and data qality. Althogh the technology is idely available, there is an ongoing need for information abot the many hardare and softare choices. This paper describes the se of data acqisition systems for compter-atomated geotechnical testing. To general approaches to data acqisition, centralized and device specific, are described, and the advantages and disadvantages of each approach are given, inclding their sitabilities for compter-atomated test control. Other reqirements for device servo (or feedback) control are described, particlarly softare featres that the athors have sccessflly incorporated in developing atomated eqipment. The paper also describes the athors' experiences ith atomating var- T. C. Sheahan, Department of Civil Engineering, 425 Snell Engineering Center, Northeastern University, Boston, Mass D. J. DeGroot, Department of Civil and Environmental Engineering, 38 Marston Hall, University of Massachsetts, Amherst, Mass J. T. Germaine, Department of Civil and Environmental Engineering, Room 1-343, Massachsetts Institte of Technology, Cambridge, Mass ios testing eqipment, and data that illstrate the qality of test control are presented. APPROACHES TO DATA ACQUISIDON There are to basic approaches to compter-based data acqisition systems: centralized and device specific. Sheahan et al. (1) have presented a detailed gide for selecting and implementing a system, inclding the to predominant methods for analog-todigital (ND) conversion, sccessive approximation and integration. They conclded that althogh sccessive approximation is the more common and cheaper of the to methods, integration is the preferred method since sccessive approximation is more likely to sample a signal aberration. In either case, higher conversion resoltion allos a larger range of otpt signals to be converted ithot hardare reconfigration. A brief overvie of the to data acqisition approaches is presented here as backgrond. Table 1 smmarizes the advantages and disadvantages of both centralized and device-specific systems. Centralized System A centralized data acqisition system consists of a single compter controlling a data acqisition control nit that receives analog signals (from measrement instrmentation and excitation voltage) from a nmber of laboratory stations. The system ill have mltiple sers and ill simltaneosly monitor a nmber of laboratory test devices, and it may have mltiple ser interface stations. A fixed terminal box at each laboratory station is connected to the central system throgh shielded cable. The box provides a point for connecting instrmentation and poer spplies into the system hile also serving as a poer distribtion and shielding panel. The compter controls system operation and ser interactions and may have a nmber of peripheral devices connected to it. The system softare can be highly complex and costly since it performs mltiple tasks, incldes a nmber of reliability and data integrity safegards, and may have other special featres. At present, the addition of softare and hardare for centralized feedback control of mltiple atomated testing devices old be extraordinarily brdensome and expensive. Hoever, the feasibility of sch a scheme can be envisioned, and some laboratories. may have a small-scale version of centralized feedback control. The principal advantage of a centralized system is that it max --- imizes flexibility, data accontability, and qality control in the

2 10 TRANSPORTATION RESEARCH RECORD 1432 TABLE 1 Advantages and Disadvantages of Centralized and Device-Specific Data Acqisition Systems (1 ) SYSTEM ADVANTAGES DISADVANTAGES Centralized Increased accontability System is critical since so many sers, Better qality control of test reslts single system More flexibility to add featres and Laboratory personnel mst share expand capabilities eqipment Improved lab prodctivity Need large lab to be economical More advanced featres Softare needs to be more general Only one ND conversion nit to Softare can't be ritten by an calibrate "amater" High speed logging may crash system Higher costs Device Specific Smaller scale, good for small labs Loer costs to bring on line Can have system redndancy ithin lab Only alternative for device atomation Can easily cstomize softare Data accontability is redced Not as many special featres in softare or hardare Becomes less economical as lab size increases Each nit mst be calibrated laboratory. In addition, only one device needs to be calibrated and tracking of system error is easier. Sch systems are economical only for relatively large laboratories, hoever, and at present cannot easily be adapted for centralized atomated test control. Device-Specific Systems A device-specific data acqisition system monitors and logs data for a single laboratory device ith a limited nmber of measrement devices. There is only one system ser and one task rnning at a time. The same terminal boxes and shielding scheme sed for the centralized system can be sed for the device-specific system. The system's data acqisition modle (prchased separately) may be inside the system's platform compter or it may be an external nit. The device-specific system provides an affordable means for the small-scale laboratory to bring compter-based data acqisition on-line. In most cases, compter speed is not an isse, and good se can be made of older compters that might otherise be obsolete. A major advantage of the device-specific approach is the simplicity of the reqired softare. It can sally be ritten in-hose and can easily be adapted to different applications. When the system is to be part of compter-atomated eqipment, control based on measrement instrmentation feedback is straightforard since the system is controlling only a single test process. In fact, the device-specific approach is the only one of the to approaches (centralized and device specific) that is easily adaptable for atomated testing eqipment. Despite these advantages, the device-specific approach has to notable drabacks: data accontability is redced and each data acqisition nit mst be calibrated. The effects of these drabacks mst be evalated as part of the atomated system's error analysis, as discssed in the folloing section. EVALUATION OF THE DEVICE-SPECIFIC SYSTEM Table 2 lists the sorces of error, in order of relative magnitde, that shold be evalated before sing a device-specific data acqisition system. Also listed are corresponding reasons for the error and check tests that can be performed to identify and qantify it. If the system is to be sed for commercial testing, it shold periodically be calibrated by a certified technician. The folloing paragraphs provide additional details on the items in Table 2. Noise Electronic noise is one reason for a loss of precision in the converted measrement signal, that is, the digital version of the tre analog signal (the other reason is robstness, hich is discssed in the next paragraph). Connection ires in the system act as antennas that pick p external electromagnetic interference, hich can add high-freqency voltage to the analog signal being converted. This effect can be minimized by employing proper system shielding and gronding techniqes [described by Sheahan et al. (1 )] and by sing an integrating ND converter. Robstness The data acqisition system can be sensitive to the type of transdcer being monitored and the circit being sed. The sensitivity of the system's precision to these inflences is knon as the robstness; the more robst the system, the less sensitive it is to these effects. For example, the system's precision may drop hen a linear voltage displacement transdcer (LVDT) is connected to it; the system may not be sfficiently robst to flly resist the effects of alternating crrent in the LVDT.

3 Sheahan et al. 11 TABLE 2 Error Evalation Smmary for Device-Specific Data Acqisition Systems ERROR SOURCE CAUSES CHECK TEST Noise Electromagnetic interference Shnt circit and monitor otpt Improper gronding Robstness Sensitivity of system to type of Verify system performance ith each transdcer and iring transdcer type Voltage Variations in analog circits for Adjst device hile monitoring stable Offset different transdcers signal Imprecise Amplification Drift Compter Clock Inaccracy Variations in semicondctors Circit losses Improperly set devices Ambient temperatre changes Circit temperatre changes Poor qality circit Poor qality compter clock Circit delays Check for increasing difference beteen measred and inpt signal as inpt increases Converted signal vs. temperatre crve. Usally random occrrence Verify ith lab timer Voltage Offset (Bias) Voltage offset is the difference beteen the inpt signal and the converted, digital representation of that signal. This offset remains constant over the opeq1ting range of the transdcer and is de to variations in the analog circit that carries the transdcer signal to the ND modle. Most systems do not have the advanced featres necessary to atomatically eliminate the bias. Ths, each time that a ne transdcer is connected to a channel, it may be necessary to eliminate the offset manally. Sheahan and Germaine (2) describe an integrating ND conversion board in hich zero offsets are corrected manally by sing onboard trimming potentiometers. Imprecise Amplification Variations in circits, voltage losses in the analog path, and im-: properly set amplification devices can case signal. amplification errors. The reslt is increasing differences beteen analog inpt and converted signal vales over the transdcer's operating range. Proper system calibration and diagnostic checks can prevent these types of errors. - Drift Drift is the difference beteen analog inpt and converted signal vales becase of temperatre changes, heating of the circit, and poor-qality circit components or circit design. The first to cases are groped nder the category of. thermal drift. Thermal drift is somehat systematic in natre, bt it is transient and cannot be corrected for; it is better avoided throgh the se of laboratory temperatre control and good circit design and hardare. Other types of drift are random and also cannot be corrected for. Timer Inaccracy Most systems mst rely on the host compter clock for determining hen readings are taken and for compting the elapsed time that shold be marked in the data file. These compter clocks can be inaccrate, and their accracies shold be verified against a reliable reference. SERVO CONTROL ALGORITHMS As previosly noted, in many compter-atomated eqipment applications, the control softare needs to be ritten in-hose. Cre,.. ating the strctre and shell for this softare is tedios, bt it is not difficlt. Hoever, the control softare's real essence, the feedback or servo control algorithm, reqires an nderstanding of the system dynamics and knoledge of available options. As Figre 1 shos, the steps in the control algorithm are collection of Digital signals read by PC Compte error in test time history Convert error to electro/mechanical device commands Commands sent to electro/mechanical devices via interface ~ardare Analog -to-digital Conversion Electro/mechanical devices actated to adjst forces, stresses & deformations Transdcers sense tre specimen state Analog signals sent to converter Ne forces, stresses & deformations applied to specimen FIGURE 1 Control feedback loop sed in compter-atomated testing.

4 12 a set of measrements, comparison of the existing state of the test specimen ith a preprogrammed state-verss-time schedle, determination of hat actions are necessary to close the gap beteen actal and schedled states, and sending ot the commands to impl~ment the action. Like any ell-organized compter program, a sccessfl servo control algorithm has many sbrotines driven from a main program seqence. Instrmentation Scan and Calclation of Crrent State The system mst qickly scan all instrmentation channels needed for servo control to captre the state of the specimen at a point in time. There shold not be a large time gap beteen measrements that ill be sed in the same control cycle or stored together as simltaneos readings (e.g., specimen deformation and applied load). To convert a particlar analog inpt signal, most device-specific data acqisition systems reqire either-a preritten sbprogram to be called from ithin the main control program or some set of command lines to be exected in the main program. To prevent a false reading (i.e~, one de to system error) from being sed as the basis for feedback control, to sets of readings can be taken and compared. If the to sets of readings have a difference exceeding some tolerance level, another set of readings is taken ntil the tolerance criterion is met. Frther error trapping may be necessary hen corrective action is implemented. To expedite the measrement instrmentation scan, the excitation voltage needs to be read only every 20 to 50 scans, provided that a high-qality poer spply is being sed. Scanning freqency can be redced for other measrements that are not expected to vary at a high rate. A check needs to be made at the end of every scan to see if the crrent set of readings is to be stored. If so, the crrent voltage readings shold be ritten to disk immediately as ra data, ithot maniplation. This ill ease error tracking. At this point in the measrement scan section, transdcer voltage readings shold be converted into the engineering nits reqired for feedback. Crrent readings can be displayed on-screen; hoever, sch displays shold be minimized becase of the relatively long time needed to access otpt devices. Determining Corrective Actions The control algorithm is intended to qickly and smoothly close the gap beteen the present state of the test process and that schedled in the programmed time history. Like nmerical analysis, control algorithms rely on diseqilibrim to fnction, ith the goal being to minimize that diseqilibrim. The error beteen the actal and target states is sed to determine the mechanical process necessary at the test device to close the gap and achieve eqilibrim. The mechanical process is actated by otpt signals from the compter to an electromechanical interface. There are three components of control: proportional control relies on the crrent difference beteen the actal and the target states, integration ses the sm of differences beteen the to states dring previos control cycles, and derivative depends on the rate of change of the difference dring previos cycles. The choice of hich components to se in the control algorithm depends on the system's response to adjstments in pressre- and TRANSPORTATION RESEARCH RECORD 1432 load-controlling devices like flo pmps, reglating valves, and mechanical load frames and on the soil's response to stress and deformation changes. 1\vo common types of control algorithms, proportional and proportional pls integral pls derivative (PID), ill be discssed. Either type of control algorithm sally reqires trial-and-error selection of algorithm parameters, althogh some manfactrers of controlling devices inclde bilt-in control algorithms or atomatic parameter determinations. Proportional Control The proportional control algorithm assmes that the error at any time, e(t), beteen the present and targeted test process states is proportional to some "stiffness" of the control system. The otpt from the compter at time t, (t), sed to correct the error is directly related to e(t). This otpt may be a signal to control a direct crrent (de) servomotor or some nmber of motor steps of a stepper motor. The governing eqation for linear proportional control is here KP is the proportional control constant or gain. This implies that the ratio of the change in pressre or load to some amont of pmp, valve, or load frame movement is a constant. The actal stiffness or gain of the system mst match that estimated in the control algorithm (the virtal stiffness). Sheahan and Germaine (2) illstrated the effects of mismatching the actal and virtal stiffness. When the virtal stiffness is too high (system is less stiff than estimated), the system behaves slggishly; hen the virtal stiffness is too lo, the system responds more than estimated and can become nstable. The athors' experience ith proportional control is ith flo pmps and orm-gear mechanisms sed in qasistatic tests that are driven by stepper motors or de servomotors. Sch tests permit slggish control behavior ithot the loss of test control qality. PID Control The PID control algorithm is fndamentally different from a proportional control algorithm becase it makes se of the system's recent performance history. For many applications, PID algorithms can achieve both a high degree of steady-state accracy and a reasonable transient response (3). 1\vo terms are added to the proportional control eqation (Eqation 1) to incorporate system performance history (or memory): an integral term and a derivative term, leading to the folloing eqation: J [ de(t)] (t) = KPe(t) + K; e(t) dt + Kd dt (2) here (t) = process otpt (i.e., control signal or nmber of motor steps sent to correct the error); e(t) = error term, hich is the difference beteen the target vale of the control feedback measrement, r(t), and its crrent vale, y(t); and KP, Ki, Kd = proportional, integral, and derivative gain factors, respectively. (1)

5 ;_ ~. Sheahan et al. Eqation 2 is a continos analog controller that is converted to a digital controller (in hich readings are taken at discrete times) by a difference eqation. For digital control applications, this type of eqation is solved by sing a z-transformation, hich is similar in principle to the soltion of differential eqations by sing the Laplace transform. For a PID controller, each comptation of the digital form of Eqation 2 typically incorporates the crrent and the to most recent vales of r(t) and y(t) and the to most recent vales of the process otpt, (t). The PID algorithm gain factors are normally evalated by a trial-and-error procedre that balances the controller's response time ith its stability. A PID control algorithm shold also inclde program statements specifying controller otpt limits. These limits, hich accont for sprios vales compted by the controller becase of noise, electrical srges, and so on, are sed henever the corresponding algorithm otpts exceed the limits. DeGroot et al. ( 4) provide other details on PID controller implementation for geotechnical laboratory testing. Sending Commands To Implement Action Any nmber of methods are available for actating pmps, valves, or mechanical load frames. To poplar types are stepper motors, hich rotate in small, discrete increments, and de servomotors, hich rotate smoothly at a velocity proportional to an inpt voltage. Both types of motors are available ith sophisticated control hardare and preprogrammed softare rotines; many motors are more rdimentary and reqire that relatively simple softare be ritten for their se. Additional algorithms for error trapping can occr dring this phase. Limits can be placed on the amont of corrective action to be carried ot by the motors to redce the impact of an erroneos measrement scan. OTHER SOFTWARE REQUIREMENTS In addition to the control algorithm and related softare, the se of device-specific data acqisition for atomated testing reqires other softare considerations. Systems can inclde a provision for restarting the test after the softare has crashed or a poer otage has occrred. At the least, the system shold be designed so that motors ill not be left rnning ncontrolled after a crash has occrred. If possible, physical ctoff sitches shold be added to prevent damage to the system. With each instrmentation scan, the program shold check hether that set of readings shold be saved; if not, they are sed for that control cycle and effectively discarded. The interval for data logging can be based on the elapsed time beteen readings or on tests of data significance. In the simplest time-based scheme, the ser enters a data logging increment, At 10 g, and an initial set of readings is stored at time tsiore After each instrmentation scan at time t, the program compares t - tsiore to At1og If t - tsiore ~ At 10 g, then the crrent measrements are immediately ritten to disk and tsiore is reset to the crrent time t. Note that readings shold not be saved in a volatile memory bffer and then ritten to disk later on; riting to disk at the time of the readings prevents inadvertent data loss. An interrpt needs to be incorporated in the program to change the logging increment, At 10 g. Data logging can also be programmed to occr according to more specialized time seqences, for example, sch1 dles based on linear or logarithmic fnctions of time. Logging increments based on tests of data significance are more sophisticated and more specific to a particlar test type. A check is made to see ho mch a particlar measred parameter has changed since the last stored vale. When the change meets some significance level, a ne reading is stored. Another featre that can be. incorporated is selective logging. At any time dring the test, the ser can tell the compter to take a reading at that point. Pressing a fnction key on the compte triggers a flag in the control loop, hich cases a set of readings to be taken and ritten to disk; the flag is then reset. EXAMPLES OF AUTOMATED LABORATORY TESTING Direct Simple Shear Testing The Geonor direct simple shear (DSS) devices (5) at the Massachsetts Institte of Technology (MIT) and the University of Massachsetts at Amherst (UMass-Amherst) se a 35-cm 2 circlar specimen trimmed into a ire-reinforced rbber membrane to an approximate height of 2.3 cm. The membrane laterally confines the specimen, providing one-dimensional [coefficient of earth pressre at rest (~)] consolidation conditions ith dead eights applied to a lever arm system. The device and the membrane lead to an applied simple shear strain deformation mode dring drained or ndrained shear. The shearing phase of the test is typically condcted at a constant rate of strain. Undrained soil behavior is stdied ith this device by condcting drained, constant-volme tests. Since the membrane prevents lateral soil deformation, constant volme conditions can be achieved hen a constant specimen height is maintained; this is done by varying the applied vertical force (and hence stress) dring shear. The changes in vertical stress reqired to keep the height constant are assmed to be eqal to the pore pressre that old develop if the test as trly ndrained (i.e., sealed) ith pore pressre measrement taken. Dyvik et al. (6) shoed the validity of this assmption for normally consolidated clay. Three generations of closed-loop feedback servo control systems have been sed at" MIT for atomated DSS testing ith the Geonor device. The first system consists of a de analog motor, an LVDT (or a proximity sensor), and a control box (Figre 2). The Contereight Vertical Load Lever Arm 13 Worm Scre Direct Crrent Motor FIGURE 2 Constant-height servo control system for Geonor DSS (4).

6 14 LVDT monitors movements in the top loading platen, hich is connected to the specimen top cap. The control box sends the difference (i.e., voltage error) beteen the LVDT reading and a reference voltage throgh an amplifier to drive the orm-gear motor at a rate proportional to the error. The orm-gear mechanism moves a lever arm, hich applies the vertical force to the specimen. This closed-loop analog feedback system cannot correct for the apparats compressibility dring shear. Data readings dring the test's shearing phase are taken by a centralized data acqisition system. The second-generation system, hich as first described by Malek (7), is similar to the system shon in Figre 2. The consolidation phase is still performed by sing manal, incremental loadings. Hoever, the control box as replaced ith a compter, an ND card, and a digital-to-anafog (DIA) card, hich are sed to read the constant-height LVDT voltage and send command signals to the de motor driver dring ndrained shear. The advantages of this system are that the control algorithm, hich is softare rather than hardare driven (as in the first-generation system), can correct for apparats compliance dring shear and data readings can be logged by. the device-specific system, eliminating the need for a centralized data acqisition system. In these to systems, only the shear phase of the test is atomated, and the shear phase of the test consists of a proportional control algorithm for servo control. In the third-generation system, a PID control algorithm as sed to atomate both the consolidation and the ndrained shear phases of the DSS test. The consolidation phase as converted to a constant rate of strain (CRS) load application. The reslting system provides more precise and detailed data dring consolidation and dramatically redces the elapsed time and the amont of labor reqired to perform a test. The PID algorithm also overcomes an inherent problem ith the se of proportional control in the DSS. The Geonor DSS ses a lever arm ith a doble flcrm system to transfer the vertical load to a specimen (Figre 2). For forces of less than 525 N, the lever arm ses only a loer flcrm for spport; for forces of greater than 525 N, it also ses an pper flcrm. When the vertical force approaches 525 N, the lever arm displaces on the to flcrms ithot casing any change in force; this transition point is knon as the dead zone. For normal incremental loadings this does not present any particlar difficlties becase the system is force controlled. It is problematic hen condcting CRS consolidation ith a proportional control algorithm becase the system is inherently displacement controlled. When approaching the dead zone dring a test, Walbam (8) fond it necessary to interrpt the compter program and manally control the servomotor ntil the dead zone had been passed. By sitching to a PID control algorithm, Ortega (9) dramatically redced the dead zone control problem and obtained excellent reslts. Figre 3 shos a typical stress-strain crve obtained dring CRS consolidation in the _Geonor DSS on resedimented Boston ble clay (BBC). Figre 3 reveals only a slight discontinity at approximately 150 kpa (i.e., 525 N acting on a 35-cm 2 specimen), hich corresponds to the dead zone. Except dring the early stages of the test, the cmlative strain rate achieved as close to the target vale of 1 percent/hr dring consolidation; hoever, the incremental strain rate as not as stable. Within approximately 40 min the PID control algorithm maintained a constant cmlative rate of strain ithin ±0.01 percent of the target vale. c > c ~ en ca :e 41 > TRANSPORTATION RESEARCH RECORD Vertical Stress cr'v (kpa) 1000 FIGURE 3 Compression crve from CRS consolidation in DSS apparats (9). Compter-Atomated Triaxial Apparats A compter-atomated triaxial apparats as developed at MIT by the first athor to perform an extensive series of K 0 consolidatedndrained shear tests (10). Its capabilities have since been expanded at Northeastern University (NU) to perform Ko consolidated-drained linear stress path tests. In addition, the improved system at NU performs its on device-specific data acqisition and logging, rather than relying on a centralized system, as it did at MIT. The system consists of the triaxial cell, a load frame, to pressre-volme controllers [PVCs, hich are based on the flo pmp design (11 )], electromechanical interface eqipment, instrmentation, and the compter ith the data acqisition card (Figre 4). The load frame and J>VCs are actated by sing stepper motors; softare commands are sed to move each of the three motors a specific nmber of steps dring each feedback control cycle. Only proportional control (Eqation 1) has been sed in this apparats, and this has been sfficient to achieve the desired control qality in the qasistatic Analog -to-digital Converter Digital -to -Analog Converter (Translator) CD Axial Force Transdcer Axial Displacement Pore Pressre Transdcer Volme Change Displacement Cell Pressre Stepper Motor FIGURE 4 Compter-atomated triaxial system. Load Frame

7 Sheahan et al. tests performed ith the apparats. Althogh all measrement devices are read so that readings can be stored, the measrements sed for feedback control depend on the test type. For types of control have been reqired in the testing programs that have sed the atomated triaxial cell. Pressre Control Dring virtally all test phases, hydralic pressres (cell and back pressres) mst be controlled. The compter takes readings from the cell and back pressre transdcers (and their excitation voltages) and converts these readings into engineering nits. In the softare, the errors are compted beteen measred pressres arid their respective target vales. Each PVC (cell and back pressre) has its on predetermined gain vale that converts the error for that pressre into the nmber of motor steps that mst be actated at each PVC to correct the error. Ths, Eqation 1 becomes CELLSTEPS = CELLGAIN*CELLERROR BACKSTEPS = BACKGAIN*BACKERROR here CELLGAIN and BACKGAIN are the respective gain vales in the cell pressre and back pressre PVCs, and CELL ERROR and BACKERROR are the respective errors beteen the measred and targeted pressres. Becase the tests are qasistatic, gain vales can be set intentionally lo, casing the system to be slightly slggish and reqiring to to three control cycles to reach targeted vales. In both long-term and short-term tests, the cell pressre remains constant to ±0.3 to 0.4 kpa [standard deviation (SD)]. This type of precise control has permitted sccessfl implementation of Ko consolidated-drained linear stress path testing at NU. Axial Load Control For tests reqiring axial stress control, the system mst be able to control the axial force applied to the specimen accrately. The compter obtains the force transdcer reading and comptes the net axial load on the specimen (after applying corrections). After determining the error beteen this and the targeted net axial load, the nmber of steps is compted for the load frame motor to move by sing an expression similar to that given in Eqation 3. Sheahan and Germaine (2) shoed the reslts from Ko consolidatedndrained creep tests in hich specified shear stress levels (q), here q = 1/i(a1 - cr3) (a1 and cr3 are major and minor principal stresses, respectively), ere maintained to ±0.1 to 0.3 kpa (SD). The ability of the system to accrately control axial load and cell and back pressres allos high-qality stress-controlled tests to be performed. Axial Strain Rate Control Since the atomated triaxial apparats as originally sed to stdy strain rate effects on soil behavior, the system needs to apply a variety of strain rates, each remaining constant dring a test. The system's strain rate control is open loop, that is, it does not rely on feedback measrements. Calibration of the stepper motoractated load frame provided a consistent, linear relation beteen (3) motor steps and axial deformation of the specimen. This calibration is sed ith the compter's clock to determine the nmber of load frame motor steps needed per cycle for a particlar strain rate. Figre 5(a) and (b) shos ho the incremental axial strain rate, deal dt, varies dring tests performed at different specified strain rates. The instability early in the varios tests reslts from to different phenomena. In the fast-rate tests [Figre 5(a)] the instability is de to initial loading system compressibility; in the slo tests it is de to the algorithm sed, hich prodces a more accrate strain rate ith time. Axial Strain Control In relaxation tests, it is necessary to control the axial strain level being imposed on the specimen. This as perhaps one of the easiest control problems to solve for this apparats since the nmber of load frame motor steps is directly proportional to the error in axial deformation (obtained from the same calibration sed for c... ~ ""-../..., <J... c. <J I- a:::: < ~ z..., <J... c. <J I < a::::. a:::: z r " {a) FAST RATES COMPRESSION EXTENSION COMPRESSION EXTENSION TIME, t (sec) {b) SLOW RATES '--~~~. ~~~..._~~~, ~~---'----~~~ TIME, 1000 (min) FIGURE S Incremental strain rate stability ith time dring conventional Ko consolidated-ndrained tests sing the compteratomated triaxial system: (a) fast rates; (b) slo rates (2).

8 16 TRANSPORI'ATION RESEARCH RECORD 1432 axial strain rate control). The reslts presented by Sheahan and Germaine (2) from to relaxation phases of a test shoed that strain levels can be maintained to :!::0.005 percent axial strain. Mltidirectional Direct Simple Shear Apparats DeGroot et al. (12) describe the mltidirectional direct simple shear (MDSS) apparats that as developed to experimentally simlate, at the element level, the stress conditions ithin the fondation soil of an offshore Arctic gravity strctre. The MDSS ses a specimen of the same size and the Geonor ire-reinforced rbber membrane sed by MIT and UMass-Arnherst in the Geonor DSS apparats. The MDSS is capable of applying a vertical stress and to independent horizontal shear stresses to a circlar soil specimen. The vertical stress (CTv) and first horizontal shear stress (T1) represent the stresses oing to the initial set don of an Arctic strctre and are applied by sing pnematic cylinders. The second horiz.ontal shear stress (T 2 ) represents the horizontal ice loading on the strctre and is applied by a variable-speed gear drive system at a constant rate of strain. The MDSS tests soil specimens nder simple shear strain conditions and is capable of applying T 1 at an angle relative to T2 ranging from 0 to 150 degrees. For a conventional K 0 consolidated-ndrained test, the specimen is simply consolidated nder the vertical stress and is then sheared ndrained by application of the second shear. Data acqisition and servo control in the MDSS is performed by sing a device-specific system based on an International Bsiness Machines-compatible PC and softare ritten in BASIC. Pnematic cylinders, controlled by voltage otpts to electropnematic (E/P) reglators, apply CTv and T 1 to the specimen. The E/P reglator maintains air pressre proportional to an electrical signal generated by a D/A card onboard the compter. Application of CTv and T1 to the specimen is atomated by sing a stresscontrolled closed-loop system ith a proportional control algorithm. Estimates of the target CTv and T 1 are applied to the specimen. The compter then collects consolidation data readings (i.e., vales of the vertical and shear strains) and also monitors CTv and T 1 via the vertical load cell and a pressre transdcer, respectively. Errors are corrected by sing a proportional control algorithm that adjsts the signals being sent to the reglators. This algorithm maintains the stresses acting on the specimen to ithin 1 kpa of the reqired level (4). As in the Geonor DSS, ndrained shear in the MDSS is achieved by maintaining a constant specimen height dring shear. A PID control algorithm monitors the specimen height ith an LVDT; a specimen height change cases the compter to actate the pressre reglator that controls CT v. Proportional control maintained the height of the specimen dring constant-volme shear, bt not ithot significant instability in CT v. Figre 6 shos the variation in CT v (normalized by the preshear consolidation stress CT~c) verss shear strain dring shear in to tests on the same soil specimen. In the first test (filled symbols), proportional control as sed and the oscillation abot the mean is apparent. The reslts of the second test (open symbols) sho the improvement in control qality hen the PID control algorithm as sed. Figre 7 plots the constant-height LVDT signal error verss shear strain dring a typical conventional K 0 ndrained test rn in the MDSS. The signal error, e(t), is compted as the difference beteen the feedback signal, y(t), and. the reference vale, r(t), e(t) = y(t) - r(t) (4) - b Proportional Controller (Eq. 1) o PID Controller (Eq.2) 1.oa--,----.~,--,---,~-,--,...--.~.,.-----r---.~.,.---,...--,-----, 0.8 "> b vi (I) ~ 0.6 ci5 co.;::; > "O 0.4 Q).!:::! co E 0 z '----'----'-~"---'---'-~"---'----'-~L----l,_-l..~.l L J J Shear Strain, 7 (%) FIGURE 6 Normalized vertical stress (avfa~c) verss shear strain from ndrained DSS tests on normally consolidated BBC sing proportional and PID controller (4). If e(t) is positive, then the specimen is being compressed; the reverse holds if e(t) is negative. Figre 7 shos that the maximm error in specimen height dring shear as less than 0. 7 µm, or percent of the specimen height. The average error dring a test as typically mch less, :!::0.1 µm, or :!:: percent of the specimen height. The maximm deviation of the constant-height LVDT's position from its target vale alays occrs at the start ~ Positive Error Eqals Comp.ression of Specimen ( 1µmeter = 2mVolt) , ~-.---,-~.---r--.-~ , ~-.---r-~r E.3- e.e...\ \..-. Cl 'Q) I +" c ~ (I) c '-----L----'-~-'------'----'----'--'--~"----'---'-~-'--'-~'-----'-----' Shear Strain, -y (%) FIGURE 7 Constant-height LVDT error dring ndrained DSS test on normally consolidated BBC specimen sing PID controller (4).

9 Sheahan et al. of the test's ndrained shear phase. When ndrained shear is initiated, the system mst qickly adjst the vertical stress becase of the high rate of pore pressre change or else lag behind in its response. As pore pressre changes become more gradal, the control system can easily keep p. A faster control cycle old alleviate the early control problems. SUMMARY AND CONCLUSIONS The se of device-specific data acqisition systems for compteratomated geotechnical laboratory testing as described. Details on evalating errors in the device-specific system ere given. Control algorithms ere explained, inclding detailed descriptions of to specific algorithms, proportional and PID. Some other softare reqirements for atomated control ere given. Examples of atomated laboratory testing ere presented to sho the control qality achieved by sing varios hardare and control algorithms. REFERENCES 1. Sheahan, T. C., J. T. Germaine, and D. J. DeGroot. Laboratory Data Acqisition: A Practical Gide to System Selection and Implementation. Advances in Site Characterization, Data Acqisition, Data Management, and Data Interpretation, Geotechnical Special Pblication 37. ASCE, Ne York, 1993, pp Sheahan, T. C., and J. T. Germaine. Compter Atomation of Conventional Triaxial Eqipment. ASTM Geotechnical Testing Jornal, Vol. 15, No. 4, 1992, pp Dorfe, R. C. Modern Control Systems. Addison-Wesley Co., Cambridge, Mass., DeGroot, D. J., J. T. Germaine, and R. Gedney. An Atomated Electropnematic Control System for Direct Simple Shear Testing. ASTM Geotechnical Testing Jornal, Vol. 14, No. 4, 1991, pp Bjerrm, L., and A. Landva. Direct Simple Shear Tests on Noregian Qick Clay, Geotechniqe, Vol. 16, No. 1, 1966, pp Dyvik, R., T. Berre, S. Lacasse, and B. Raadim. Comparison of Trly Undrained and Constant Volme Direct Simple Shear Tests. Geotechniqe, Vol. 37, No. 1, 1987, pp Malek, A. Cyclic Behavior of Clay in Undrained Simple Shearing and Application to Offshore Tension Piles. Sc.D. thesis. Massachsetts Institte of Technology, Cambridge, Walbam, M. Procedre for Investigation of Sample Distrbance Using the Direct Simple Shear Apparats. Master's thesis. Massachsetts Institte of Technology, Cambridge, Ortega, 0. J. Compter Atomation of the Consolidated-Undrained Direct Simple Shear Test. Master's thesis. Massachsetts Institte of Technology, Cambridge, Sheahan, T. C. An Experimental Stdy of the Time-Dependent Undrained Shear Behavior of Resedimented Clay Using Atomated Stress Path Triaxial Eqipment. Sc.D. thesis. Massachsetts Institte of Technology, Cambridge, Olsen, H. W., R. W. Nichols, and T. L. Rice. Lo Gradient Permeability Measrements in a Triaxial System. Geotechniqe, Vol. 35, No. 2, 1985, pp DeGroot, D. J., J. T. Germaine, and, C. C. Ladd. The Mltidirectional Direct Simple Shear Apparats. ASTM Geotechnical Testing Jornal (in press). 17