Raymond E. Dessy Chemistry Department Virginia Polytechnic Institute Blacksburg, Va. 24061
The art of analytical chemistry is being altered by the impact of computer technology. At one extreme, large data bases are being correlated to abstract more information from existing analyses, and these results are being reported in a format more adaptable to human senses by means of sophisticated graphics software and hardware. At the other extreme, tasks involving large numbers of repetitive operations or in which large amounts of data are made available by each experiment have been successfully automated. The implication of this last word is NOT just that some computing element has been added to the system, but that the manual operations involved in instrument setup, calibration, sampling, and data collection have also been relegated to an electromechanical servant. Lab automation involves much more than just acquiring a computer and also more than collecting data from the instrument automatically. This article analyzes the broad themes that have dominated the last two decades of lab automation and comments on the wave of the future that will correct existing or rapidly developing deficiencies in most lab automation schemes. During the 1950's it became obvious that the maxicomputer which was ubiquitously solving the problems of the business area of corporate work would find application in the research laboratory. In the succeeding decades, systems analysts, often not chemists or chemical engineers, extended the classic computer science disciplines to the correlation of data from the quality control, process control, and analytical laboratory. Several instances of real-time participation can be cited, but most of the emphasis was on terminal and key punch data entry. Often political structures developed around these computers which effectively discouraged real-time laboratory automation schemes. And today's laboratory manager or research director often matured in this maxicomputer environment. In the 1960's the minicomputer emerged as a tool in laboratory automation. It provided some of the abilities of the large mainframe computer in addition to excelling at the input/output operations so necessary in an environment in which real-time data acquisition and control are involved. In the 1970's the new thrust in automation is found in the area of microprocessors and microcomputers. These large-
scale integrated (LSI) circuit devices proffer very inexpensive routes to control and data acquisition, prompting many instrument designers to incorporate them in classical laboratory equipment, thus introducing the age of the smart instrument—an analytical tool that is partially automated as delivered and contains elements intelligent enough to perform the tasks of autocalibration, data collection, and transmission for the user. The mid-1970's have also seen the development of computers that lie between the maxicomputer and minicomputer, the midicomputer. But there are still shadows on the landscape. In many laboratories lab automation has been nucleated by a minicomputer serving several instruments, thus creating small islands of computer capability. These have grown to a critical and optimum size—and then frozen in a fixed configuration. Financial and real-time constraints forbid an extension of their capabilities or resources. These systems continue to consume maintenance funds until amortization permits replacement. Is there an alternate expansion route? Increasingly, chemists frustrated by the lack of response to their automation needs are seeking redress in intelligent instruments. This will improve sample thruput, thus accommodating the current pressure on most laboratories. But record keeping and report writing will increase. Computing center personnel are entranced by the data base management programs such as Image, Total, Mumps, IAS, etc., which promise to meet today's increasing fiscal and legal imperatives. The knowledgeable lab manager sees the opportunity in these programs for correlative and retrieval mechanisms that will answer his manpower problems. And the statistics conscious bench chemist would welcome computer-assisted tracking of equipment performance. However, this increased data flow demands that the extremes of computing technology, such as data base management and input/output (I/O) control, be brought together so that they can interact. A route to open-ended design for laboratory automation is evolving to permit computing facilities to share resources, capabilities, and data. Such a scheme allows small systems, with just sufficient hardware to accomplish a specific task, to be located near the experiments or chemical process. Yet this
1100 A · ANALYTICAL CHEMISTRY, VOL. 49, NO. 13, NOVEMBER 1977
Report
COMPUTER NETWORKING A RATIONAL APPROACH TOLABAUTOMATION small system can call on and use all of the powerful facilities associated with larger computer systems during program development and debugging phases, including the mass storage elements needed for retrieval and correlation of large data bases accumulated via multi-instrument analytical services. Routine reports can be computer generated. This route is now becoming available in what is termed networking or distributed processing. The purpose of this article is also to explore the strengths of networking and of the various tools used in automation to emphasize how one can simultaneously maximize the strengths and minimize the weaknesses inherent in each type of tool by uniting them in an intercommunicating, synergistic manner. Viewed as separate entities, the three classes of automation tools are remarkably complementary, even to the casual observer. • The larger midicomputers are systems that operate under the control of exceedingly complex software systems—called operating systems. They have excellent computing power (ability to manipulate vast amounts of data in complex ways) and can support vast amounts of memory, either as fast semiconductor or core memory, or in the form of'rotating magnetic storage media called disks. Expensive printers and plotters can be added to communicate effectively with the users who can justify their existence. These users may be laboratory managers who need access to distilled progress reports on the nature, amount, and quality of the data their analytical groups are producing, or supervisors who need exposure to the financial and efficiency factors so important to proper system growth. • Minicomputers operate under the control of much less sophisticated operating systems. Operating with minimum overhead in executing software, they can respond rapidly to the needs of their instruments; operating without large numbers of attending systems engineers and analysts, they can readily respond to the needs of their user. • Microcomputers are the offspring of the large-scale integrated (LSI) circuit technology that gave us hand-held calculators, and the electronic equipment that dominates the point-of-sale market, and are invading the household entertainment and appliance areas. These latter implementations
require inexpensive devices that can control motors, heaters, or valves; that can take data from push buttons, thumb-wheel switches, or joy-sticks; and that can display data in lights, panel displays, or on TV tubes. The calculations required are minimal, and the need to expand is nonexistent. These features have made them the designer's choice in creating intelligent instruments. Since laboratories have devised their own schemes to implement automation, all of these hardware tools mentioned have been employed individually. Each has its own advantages and limitations, usually discovered idiosyncratically by individual users, who have learned to cope with the weaknesses. Midicomputers incur such large software execution overheads that they cannot respond well to the real-time needs of a large laboratory. As the single agent for automation, the highly interactive nature of software tasks serving multiple instruments makes software generation and debugging of the midicomputer system a complex problem, far outside the capability of the average scientist. The intelligent instruments require programs that only need to take cognizance of a single experiment; however, developing these programs on a small CPU involves inefficient use of manpower. Even minicomputers lack the storage and speed capabilities needed to perform the cross-correlating and concentrating functions on large data bases. In many laboratories where separate development of automation schemes involving two or three of the above routes is on-going, panic is developing because of the limited or nonintercommunication between isolated systems. But in the several laboratories where responsible personnel are technologically competent and mutually cooperative, the synergistic interaction of micro, mini, and midicomputer facilities has led to the development of an intercommunicating network of automated facilities. Classic examples of successful networking can be seen in industrial process control facilities such as Union Carbide or Philip Morris, and in academic laboratories such as the University of Oregon. However, the inaccessibility of software or nontransportability of software has precluded widespread dissemination of the techniques.
ANALYTICAL CHEMISTRY, VOL. 49, NO. 13, NOVEMBER 1977 · 1101 A
Major changes in the marketing strategy of the large computer vendors are rapidly altering this state of affairs. They have analyzed the computer markets that are rapidly saturating or becoming excessively competitive, and have developed an increased awareness of end-user needs. As a result, these vendors are beginning to provide almost turn-key network hardware/software packages in a variety of forms. From its classical position as a maxicomputer vendor, IBM is moving to serve the laboratory market with its introduction of Series 1 and the Device Coupler in the mini and micro areas. It undoubtedly will follow with a complete software support package using the network protocols it calls Synchronous Data Link Communication (SDLC). Hewlett-Packard, Modcomp, and Data General offer network packages. Digital Equipment Corp. with its significant position in the micro, mini, and midi area has begun to strengthen its integrated offerings via a network protocol called Digital Data Communication Message Protocol (DDCMP). And the IEEE Laboratory Bus Standard (a simplified network approach) is now commercially available in the form of the HewlettPackard Lab Bus. Many other computer vendors provide network products, and a few instrument vendors such as Spectra Physics are making important innovations. The questions of WHY?, WHAT?, and HOW? will be explored so that one can judge one's own laboratory's needs for networking capabilities. This is done, not with the anticipation that the average chemist will implement the net, but rather with the firm conviction that the laboratory personnel must be the most important factor in determining the final scenario for lab automation. To do this requires that a vocabulary and philosophy be acquired so that the chemist can have an affective and effective voice in the decision. The establishment of a network facility is indicated under a number of etiologies: • Real, not imagined, fear of total dependence on one computer • Acquisition of a number of intelligent instruments which leads to increased thruput which, in turn, leads to a data correlation and report generation bottleneck • Addition of another instrument to a multi-instrument minicomputer-based system which taxes the real-time capability of the central processing unit (CPU). Either the rate of data acquisition must be compromised, or some instruments are forced to idle at peak load. • A minicomputer system becomes so burdened with real-time data needs (which operate in the so-called foreground) that the background tasks of data correlation and report generation become backlogged and the system response degrades. Software development costs escalate. • A midicomputer system that has been successful in its role of a business-oriented machine is asked to handle the control and data acquisition needs of laboratory equipment. By adding networked micro, mini, or midicomputer capabilities, each of these scenarios can be successfully resolved. Obviously, each of these examples could also be resolved by a nonnetwork addition of computing/automation hardware. However, networking is the only route that will lead to an open-ended, expendable automation scheme—and one that can use much of the existing equipment. In it, the individual elements will be able to articulate one to another, developing the full capabilities of each member, rather than forcing one member to inadequately duplicate the ability of another. It is obvious from these examples that networks can grow either from the direction of satellite instrument or computer up, or the maxicomputer down. The elements of growth potential and intercommunication between elements are the heart of a computer network. In a microcomputer or intelligent instrument a stored program determines what actions are to be taken. The stored program has only three functions: to interact, if necessary, with the operator of the instrument to perform the analysis
correctly; to perform the data acquisition operation; and to provide a usable report to the user. In this latter process it may manipulate the data in a simple manner to provide a distilled report to the operator. The "smart" GC and LC equipment currently being vended commonly performs peak picking, sample identification, and quantification based on user input and vendor algorithms. Atomic absorption equipment likewise can establish calibration curves based on standard samples and then provide information on heavy metal content in the same samples which have been subjected to GC or LC. These programs are provided by the vendor in fixed format "burned" into PROMS. The user has little or no knowledge of their operation. But some vendors also provide the capability of transmitting either the entire or the distilled data file from the intelligent instrument to a more sophisticated computer. This will become increasingly valuable as networks develop. It is quite common to have intelligent instruments coexist in a laboratory equipped with special analytical devices automated by small computers. These require that specific programs be developed to elicit data collection from the same samples subjected to GC, LC, or AA. However, there is no simple way to correlate the output of these various instruments other than having a chemist or technician monotonously transcribe the data into a compiled list. It is possible to interconnect computer elements, so that laboratory computers can transmit their data to a larger host computer. It would then be a trivial matter for the host computer, with sufficient rotating memory, to store the output from each instrument over long periods of time, to implement searches through each list of data, and pair up matching sample numbers. A correlated report which flagged significant concentrations or relationships could be produced. In many laboratories and particularly in process control environments, the intelligent equipment near the experiment often becomes insufficient to the task. Under changing conditions, different decision processes must be used, which require different programs. Manual intervention or storing of all conceivable software in an expanded memory configuration is a possible solution. Alternatively, a hardware link from a host to a satellite would permit "down-line" loading of programs into the satellite and initiation of execution. This process can be augmented by programs running in the host which make decisions about the operation of the satellite and automatically change its programs. As experiments become more sophisticated and the time and function demands made on computerized systems increase, many analytical chemists will want more than an intelligent instrument on their bench. They would like to develop special function programs and use the facilities of a larger system for printing, plotting, and manipulating their data. In the smaller computers, software development costs represent over 80% of the installation cost. As hardware decreases in price and software increases, it becomes important that software development aids be used extensively. These tools are available on mini and midicomputers running good operating systems, and support the CRT terminals, line printers, etc., which speed programming and improve documentation standards. It is possible to create programs on these host systems in assembler, BASIC, or FORTRAN in such a way that they can be down-line loaded in machine code to a satellite and executed. In the satellite they may be executed and debugged without fear of bringing the main system down; yet, when fully operational the satellite can use all of the resources of the machines above it. Finally, each of the satellite users does not need to know how these facilities are provided. This brings us to the point of trying to understand how a network operates. The heart of a computer network is the protocol conven-
1104 A · ANALYTICAL CHEMISTRY, VOL. 49, NO. 13, NOVEMBER 1977
tions that it employs. 1 hese are the accepted conventions used to electrically and logically transmit information from one computer to another. These protocols are layered, one on another, to span the range from the hardware plugs used to interconnect units, through the software commands the user sees that implement network functions. Most chemists will see only the outer layer of the onion, but the entire core must be there. For example, a complete software support package called DECNET com pletes the DDCMP protocols to provide a user-oriented product. User command protocol—formats for user requests at a terminal Machine command protocol—formats of these requests in machine code Logical link protocol—formats that assure sequential re construction of individual message packets Physical link protocol—formats that assure error-free transmission Hardware protocol—electrical formats used on the hard ware interconnection. To more easily understand how these protocols work, imagine a commonplace operation—placing a telephone call—and you have parallels to all the operating parts of a computer network. Let's examine a situation involving transmission of a pro gram from the host computer to a satellite. The process starts hv issninir a enmmatiH from the satellite
RUN CPUl;DXO:XRAY asking that a program on CPU1, located in disk drive 0, and called XRAY be down-line loaded from the host computer to the satellite and run. The equivalent process in our telephone system would be dialing a number. The dialing process es tablishes source and destination information and initiates a sequence of electronic switching operations that interconnect the source and destination. Computer networks mimic this activity. In a computer network the user command (RUN ) is converted to equivalent machine code commands. These commands, containing source and destination information, are formed into a message packet constructed along rigid logical link protocols and sent down the wires interconnecting the devices in an attempt to create a connection between source and destination, called a "handshake". In some networks all of the members are attached to a common party line or multidrop. The destination only re sponds when a transmitted signal activates it, just like a party line telephone, and other units disregard the signal. The IEEE Laboratory Bus works this way.
Point-To-Point Network
Other networks have point-to-point connections between members, often in a star or hierarchical arrangement. This is akin to a private leased wire. DEC's software package RT11/REMOTE operates in this fashion. In very large systems, multinodal arrangements occur with many possible pathways from source to destination. It is the function of the original handshaking message to find the best available route for the communication by trying various pathways, discarding them one-by-one if they are busy. The DEC software package DECNET running under RSX-llM is an example of this sophisticated networking form. Such networks can survive failures of one of the computer members with minimal loss of effectiveness.
Multinodal Network
ΕΚΕ32Ξ SI
As the initial message packet establishes a link between facilities, a positive acknowledgment scheme is used to ensure proper operation. Each message packet sent demands an ac knowledgment from the destination before further commu nication occurs. A parallel exists in our telephonic world, ANALYTICAL CHEMISTRY, VOL. 49, NO. 13, NOVEMBER 1977 · 1105 A
Message Packets Supervisor Manager Programs, Data
To Host
Source Destination Function
Positive Acknowledge
where HELLO, GOOD MORNING, or an identifier such as a company name is used to acknowledge a connection. Once the connection is validated, the data or program must be transmitted. This involves a very complex set of logical link protocols which assure that complete messages are received and reconstructed into a conversation in proper order. A companion physical link protocol is used to assure error-free transmission. These protocols differ in form and complexity from vendor to vendor just as human languages vary from country to country. Human communication delimits thoughts by sentences, and each language has its conventions with respect to location of nouns, verbs, adjectives, etc. The computer's intercommunication protocol at this level agrees on the types of messages permitted, and their format. A typical set is: Management messages—carry command information such as load, run, read, write Supervisor messages—maintain the orderly flow of information after the management command has been transmitted and execution begun Data messages—programs and data Regardless of the type of packet being transmitted, the source waits for a positive acknowledgment before attempting further transmission. It is never safe to assume that the communication link is error free, even though primitive nets have done so. In telephonic communication, garbling of words is easily detected by the listener since the word heard does not fit a known vocabulary. Loss of words is detected by a sentence becoming meaningless. And loss of entire sentences is usually detected by a long silence. Finally, we know when words, sentences, and major thought changes occur by pauses.
The computer messages have a similar construction and similar problems. The message packets begin randomly in time, and the destination must recognize or "frame" on the start of the message. Synch characters are transmitted to alert the destination to an incoming message packet. These synch characters are different from vendor to vendor and for various hardware protocols. The message is made up of bytes of data, 8 bits of information long. Proper framing on the beginning of a byte, and on each bit is also essential. Any good networking system has methods of assuring that neither bits, bytes, nor entire messages are lost in transmission. Each message packet is numbered in sequence so the destination can detect lost messages. Each message packet contains a number representing a byte count, which can be used by the destination to detect lost bytes. And some form of telecommunication error checking, such as parity, checksumming, or cyclic redundancy check (CRC) error detection, is used to detect transmission of bad bits. If bit, byte, or message errors are encountered, the positive acknowledgment system used ensures that the message packet in question will be retransmitted. Well-designed network protocols will also accommodate the equivalent of temporary and permanent disconnects and babbling satellites.
Byte Loss
Bytes in Message Message Packet
of Message
Message Loss
01011111
01011111
11100111
11 00111 Row Error
10001000
10001000
01010101
01010101
11011101 Checksum 11 11101 Column Error
Framing Error Checking Procedures Parity: Sum of 1 's in byte should be even Checksum: Transmitted and calculated checksum should be equal
Message
CRC Algorithm 10110111 The language used in the protocols involves the transmission of both data and control characters. In the HP/IEEE Laboratory Bus the control language has been created specifically for the system. Other vendors will use American Standard Code for Information Interchange (ASCII) or Extended Binary Coded Decimal (EBCDIC). For the most flexible use, these networks should be capable of handling all possible characters in these sets—8 bits will permit 256 1106 A · ANALYTICAL CHEMISTRY, VOL. 49, NO. 13, NOVEMBER 1977
combinations. This assures that straight binary information can be transmitted. This feature is called transparency. Some simpler systems will only support transmission of alphanumeric characters (upper and lower case) plus a limited set of control characters. Finally, at the hardware protocol level we are concerned with the electrical form in which the data are transmitted. Common methods of transmitting data involve parallel or serial routes. In the former, the bits representing the data word are shipped simultaneously down multiple-conductor cable. This permits fast transfer of data or control words. The HP/IEEE Laboratory Bus is 16 bits parallel, 8 bits of control information, and 8 bits of data. This convention has many advantages, but increases cost and limits the length of the bus (20 m). Serial transmission links involve sending one bit at a time of each byte down a twisted pair of lines. This transmission can involve a synchronous link (where synch characters sent down the line preceding the message assure that the receiver and transmitter are operating concordantly, and then a burst of bytes are sent) or asynchronous links (where each byte is preceded by a START bit which tells the receiver a character will follow). Synchronous links offer speed advantages when large data bases are being transferred, but they currently engender a higher cost. Asynchronous linkages are used in many networks and are always found in intelligent instruments which offer computer compatibility. Asynchronous serial links are available in two protocols—either a 20-mA current loop or an RS-232/C voltage convention. The source and the destination must both operate with the same protocol and at the same transmission rate. Intelligent instruments and common micro and minicomputer serial ports commonly transmit from 110 bits/s to 50 kilobits/s. Most intelligent instruments meet hardware protocols, but lack the proper logical protocols. It is often necessary to inject a microcomputer between them and a network. To give the analytical chemist a feeling for the flexibility and expandability of networking, let's examine a possible development pattern in a quality control/process control/ research analytical laboratory environment common to chemical companies. Few systems would ever grow as large as our final result, but the pieces are representative building blocks for similar systems, all of which represent actual installations currently in operation in the U.S. or Sweden.
In a polymer physical testing laboratory faced with tensile strength, elongation, cold flow, and molecular weight determinations, the IEEE Laboratory Bus has some interesting advantages because scanning analog-to-digital conversion (ADC) equipment, frequency meters, voltmeters, relay actuators, and printers are available from numerous vendors with the IEEE Laboratory Bus adapters built in. These are attached to a DEC LSI-11 microcomputer via the LSI-11/ IEEE Bus interface card. The microcomputer easily handles the control/data flow and report generation in the system, the programming being done on a host DEC PDP 11/34 used for laboratory management purposes. The system operates independently of the host PDP 11/34 in case of its failure or saturation. Local reports are generated either on strip printers at the instruments or on the terminal of the LSI-11. Condensed data are sent for correlation, accounting, and backlog statistical purposes to the disks of the host PUP 11/34.
Mini
Micro IEEE Lab Bus
Polymer Lab
In a quality control laboratory, which analyzes samples from customers and from process control streams, intelligent instruments with serial ASCII RS-232/C outputs are attached to a microcomputer that allocates sample numbers and other identifying information, concentrates the data, and passes this to a larger host. This could be the same host PDP 11/34 previously described.
Mini
Micro
Customer Service
Process Quality Control Control Lab Polymer Research Lab Lab
Intelligent Instruments Quality Control Lab
ANALYTICAL CHEMISTRY, VOL. 49, NO. 13, NOVEMBER 1977 · 1107 A
Mini Midi Micro
Research Lab
Mini Micro
Vibration Experiment
Process Control
In a research laboratory, data collection from a special vibration experiment is controlled by a microcomputer subsystem netted to a host computer where the actual Fourier transform is made. This could be the same host as before. The physical-inorganic testing laboratory of a production facility must log melt characteristics over long periods. It is using the IBM Series 1 Minicomputer and Device Couplers. Device Couplers are microprocessor-based units that allow programming in A Programming Language (APL) to sequence manual steps in the analytical process, to acquire and buffer data for transmission to a host, and to accept calculated data from the host for display at local video-terminals. The Device Couplers connect to the Series 1 via asynchronous Binary Coded Decimal (BCD) links. The Series 1 will eventually be netted to the Corporate IBM 370 via Synchronous Data Link Communication (SDLC) protocol software.
Acknowledgment The research group at VPI&SU uses a network of two intercommunicating DEC PDP-11 hosts attached to four other satellite computers. This computer network has been developed by the author and M. Starling, W. Nunn, D. Hooley, H. Wohltjen, C. Knipe, D. Binkley, J. Berquist, G. Giss, E. Fiorino, J. Fiorino, C. Baker, I. Starling, and I. Bowater.
Maxi
Mini
To accommodate similar user functions described above and to provide local data base management facilities, a PDP 11/70 midicomputer serves as a host to the PDP 11/34 and as a satellite (via SDLC conventions) to the Corporate computer. Over the next few years the software to support file transfer in transparent mode between mixed vendor processors will become available. With little effort and an open mind, today's analysts will soon use the entire resources of a complex network as casually as he uses a hand-held calculator today.
Color
Unrestricted grants from the Gillette Charitable and Educational Foundation have supported this development.
Viscosity
Customer Service
Device Coupler
Atomic Absorption
And finally in a process control environment, with over 500 temperature, pressure, and flow rate sampling points and 200 chemical analyses stations, a very high-speed (1 megabit/s) coaxial line connects a DEC PDP 11/34 to several microprocessor-based industrial data acquisition stations.
Raymond E. Dessy is professor of chemistry at Virginia Polytechnic Institute and State University. He and his research group are currently involved in the design and development of automated analytical instruments, development of new detectors to permit the study of fast reaction mechanisms, laboratory automation for the small and large scientific laboratory, development of courseware, software, and hardware for teaching microcomputer and minicomputer interfacing and applications using the DEC LSI-11, and development of an audiovisual course on laboratory automation for the ACS/NSF project CEDS.
1108 A · ANALYTICAL CHEMISTRY, VOL. 49, NO. 13, NOVEMBER 1977
