Railway traction control systems

Bookmark and Share

Cecube technology consulting


This page is a resource for control systems and techniques used on modern railways, in particular rolling stock traction equipment.


When working on a traction project a frequent control question was encountered - how to effectively counteract inherent deadtime present in a control system. The system in question was a DC motor drive and in this instance it was feasible to change the hardware to reduce the deadtime problem. However, in many cases this is not possible, so what are the alternative control solutions?

Deadtime arises from a delay in propagation of a control demand to the controlled output, and during this time there is no response at the output. Common traction sources of deadtime are fluid transmission in hydraulic and pneumatic controls, backlash in a gearbox or coupling, and firing delay in a phase controlled rectifier. The significance of deadtime is calculated from the product of deadtime (in seconds) and required bandwidth (radians / second). If this dimensionless value is greater than 0.5 then additional compensation is often necessary to achieve acceptable stability and response. Additional Control System Support items are located under "Helpful Hints".

It is a common misconception that deadtime can be tuned out by appropriate derivative term within a PID controller. Derivative correction can compensate for phase lag in the controlled plant, but it is always ineffective against pure time delay. Instead the classical deadtime compensation was conceived in 1957 and is known after its originator as the Smith Predictor method. Its control structure is shown in the block diagram below. It requires a reasonably accurate model of the plant excluding deadtime and a separate series function block that estimates the cumulative deadtime delay of the plant. Provided sufficient a priori knowledge exists, the value of the delay estimate may be scheduled according to the system operating point.

Smith Predictor

By feeding back the estimated undelayed model output, the PID control is driven by a fresher signal. Only a small error signal, representing latency error, is added to finely adjust the PID response to remove closed loop differences between the model reference and the actual plant. There are many variations of the Smith Predictor scheme to alleviate sensitivity to model errors, but increasing complexity can lead to diminishing returns, and often the simplest implementations are the most effective.


Proprietary simulation packages for electronic / electrical / mechanical control system modelling support the control system design engineer. Such tools have a rightful place along side CAD/CAE mechanical engineering packages in the development cycle of traction equipment. Those familiar with rail traction modelling prior to the year 2000, or if you read Paper 3, will know there were two distinct genres of simulator control design product.

  1. Electrical and Electronic Circuit Design tools
  2. Multi-Discipline System Modelling tools

It could have been expected that these two families would merge over time, but actually the opposite has occurred, driven by the overriding influence of the large scale micro-chip manufacturers. As chip life cycles have shortened, fabrication scales reduced, and transistor component counts multiplied, so development simulation has become ever more paramount. A lucrative simulation industry has grown around silicon based Micro-Electro-Mechanical Systems (MEMS), and a set of high-end vendors compete in this market. These players include ARM, Cadence/CoWare, Coventor, Dolphin, Mentor Graphics and Synopsis. These vendors have a family of products to support the main MEMS electrical and physical modelling centerpiece. The chip simulation packages utilize the public domain SPICE engine in conjunction with other physical modelling tools such as ANSYS and Hardware Description Language (HDL). For example, Synopsys produce software tools for the design, synthesis, and validation of FPGA and ASIC integrated technology. Also in the same portfolio are packages for CAD, mixed-signal analysis and analogue simulation. The following Synopsys tools are amongst those available:

  • Analogue and Digital Simulation (combined HSPICE [SPICE augmented for 64 bit operation and high frequencies])
  • Front-end Device Design and Verification (including Verilog and VHDL mixed)
  • TCAD (micro-electronics diffusion and transport equation solver)
  • Saber System Development Modeler (for non device physical system simulation
  • Analog / Mixed-Signal (AMS) Verification unit that integrates functionality of these four components

It is irony that HSPICE is teamed with Saber as complimentary products within a suite for IC development, despite being past competitors. Given the development effort into SPICE integration tools funded by the micro-electronics industry, there is no credible alternative for circuit development - it's the industry standard. A full featured basic version called LTSPICE is available via the recommended links. The LTSPICE standard 3f5 version engine allows full numerical integration control and versatile plotting facilities with FFT spectral analysis options. The schematic capture is less intuitive than most, however, with experience is acceptable. See LTSPICE download package details for licence terms. With SPICE Windows versions available as free downloads to top end UNIX licences at over 100,000, an entry level for every organisation exists. Importantly, migration from one SPICE product to another is possible, although more difficult for the top end.

Over time SPICE behavioural modelling (representation of physical systems by programmable voltage and current sources) has developed to permit circuits to be part of larger systems. However, it is yet to become an alternative for large scale control system modelling. There is little incentive given the profitability of the micro-chip market. The control system modelling market is dominated by MATLABTM. MATLAB, from the Mathworks, solves system problems represented in dimensionless matrix form, whereas the SPICE engine solves electrical networks by Kirchoff's Current Law. Originally MATLAB just solved continuous time system equations, but has expanded with comprehensive tool-boxes covering areas from digital (z-plane) design to robust controller optimisation. Tool-boxes for power electronics and circuits as well as mechanical components exist to extend MATLAB capabilities, but complex mixed-mode electronic circuit simulations are handled less well. Like SPICE. MATLAB is the industry standard in its field. However, there are alternatives, most of which offer MATLAB compatibility or at least similarity. Matrix-X was a strong brand in the 1990's, but now VISSIM, LABVIEW and SCILAB are the main alternatives to MATLAB. VISSIM has good MATLAB compatibility and code generation facilities, but does not offer such an extensive range of control tool-boxes. LABVIEW has gained popularity in recent times, and is particularly well geared to process control and data analysis problems. SCILAB offers a terrific free product capable of solving many small or medium sized problems. It has strong UNIX (or rather GNU) origins, and is now also ported to Windows OS's offering a Java solver engine and integration platform.

So the control tools sector is a mature market. Two industry leading standards in SPICE and MATLAB have emerged. These products deserve their support for good and differing reasons. For major engineering organizations to significantly resource and fund projects based on other tools not committed to either of these standards could result in future maintenance problems.


The illustration above of deadtime removal by a Smith predictor is an example of an observer, used to determine correct feedback in this case. The state observer is a control technique used to identify unobservable quantities. This is particularly useful in control of modern traction motors. The predominance in traction motors drives is either the induction motor or the permanent magnet synchronous machine, sometimes referred to as a "brushless DC motor". Both motors are characterized by the absence of a physical connection to the rotor. In the former case the rotor flux is developed in the rotor bars by induction and in the latter the rotor field comes from rare earth magnets. This means the flux cannot be directly deduced by simple current measurement. This is a major problem since the preferred method of control is the vector principle. In vector control technology the angular position () and magnitude of the flux vector is the control variable that decouples the interaction of torque and flux producing stator current components. Unfortunately the flux is unknown, so a state observer is required to provide this information to the main controller. The diagram below shows how the observer is integrated into the feedback system.

State Observer

A good observer invokes a model of the motor that overcomes the following limitations:

Design of a robust state observer may take many forms and is a challenge that has exercised many academics and resulted in a wide range of solutions, with varying degrees of success. Two of the most recognized academics in this field are Leonhard W and Bose B K, whose early texts for further study include papers 19 and 21.

Contact Details top of page

Simulation Tools

Custom Search

Loading news - Please wait

Simulation Links

MATLAB Central

End of links