EMC and errors: searching for a quiet life in the control cabinet
Bus systems have become an integral part of machines and industrial plant. However, data transmission can be impaired by electromagnetic interference  electromagnetic (in)compatibility. The close proximity of power electronics and data lines may cause problems which can result in data errors... Electrical engineers will probably know something about the statistics of data corruption while most mechanical engineers will associate EMC with intractable and idiopathic problems. EPSG has carried out analytical test work which sheds interesting light on EMCinduced errors. Specifically the results of new work suggest that the method of network data framing has a significant effect on error rate
MODERN MACHINES include numerous electronic power components for the running and control of drives and servo motors, in particular frequency inverter and servo inverter systems. Such devices inherently produce voltage gradients which are a product of both the (high) voltage being switched, the switched current and the inverse of the switching time (rise time). One could also add to this mix the ever higher switching frequencies employed in the search for conversion efficiency. Taken altogether inverters can be an incredibly powerful and pervasive source of broadband electromagnetic interference. But of course in any industrial plant, there will be other potential interference sources as well.
Many possible sources of errors
Let us look at a typical situation: an error diagnosis system reports a controller failure, but subsequent analysis shows the suspected components to be functioning properly. It takes much time and effort to find out that the malfunction is not caused by defective parts but by electromagnetic interference. Modern diagnosis systems can locate errors more or less precisely. Practical though this is, it only hints at the location of the error, not its cause.
Maintenance experts can take their pick from a whole bunch of possible explanations: initially, it is not obvious whether they face software problems, defective hardware, bad wiring, bus faults, possibly even a rare memory error caused by cosmic radiation, or whether the cause lies in electromagnetic incompatibility of the components. Tracking down the source of the interference or the segment with insufficient shielding resembles the proverbial hunt for a needle in a haystack.
However robust they are, all machines are subject to wear, and the more they wear out, the greater becomes their susceptibility to interference. Continual (electrical and mechanical) shocks strain the contacts while and insulation (and electrical screening) of moving parts such as drag chains suffer during sustained operation. Since wear unavoidably increases vulnerability to interference, all potential weak points must be avoided in the design phase. Unsafe machines can hardly be made safe afterwards. EMC strategies need to be observed from the outset.
Comparative analysis
The communication network is the machine's nervous system and, as such, vital to its correct functioning. Since data traffic is necessarily exposed to electromagnetic interference, the transmission protocol must ensure a robust communication structure. While the destruction of some data packets cannot be prevented altogether, it must not impair the application. What follows highlights the critical mechanisms in data communication – and hopefully offers some expertise in controlling them.
Frame transmission can be organised in two ways, either using a sum frame procedure, wherein one frame supplies data to several participants, or transmitting data to each participant in an individual or single frame. EtherCAT uses the sum frame procedure. EPL and Profinet use individual frames. The difference in data framing is shown in Fig. 1. The following theoretical observations identify characteristics of both procedures.
Fig. 1. Sum frame and individual frame properties
The two major causes of electromagnetic interference come from the power supply to network devices and by electromagnetic waves coupled into data cables. For the sake of the test, both causes can be considered as one, and power supply, sender, receiver, and cabling can be regarded as a unit. To make the test results comparable, the structure was standardized and the systems were connected in line.
In a first approximation, we assumed 40 bytes of payload data. The test did not include particular effects such as I/O data asymmetry, i.e., differing amounts of input and output data at one node. One error destroys a complete frame, presuming that the errors are stochastic – as in not directly related to the network communication. We further assumed punctual effects – one error impairs maximally one frame. If n is the number of routes which are connected in line and P is the error probability for one route, then the error probability for the complete route is:
P_{total} = 1 – (1 – P_{individual})^{n}
If P is relatively small compared to n, the formula can be simplified for an approximate result. Then, the error probability is:
P_{total} = n ×P_{individual}
In the following, P will be seen as the probability of a 40 byte data block being impaired. To simplify data comparison, we suppose the overhead to also measure 40 bytes, although this is not exact. The final calculation uses the actual values.
Individual frame procedure
All individual frames are transmitted via the first route; afterwards, their number decreases by one at every node. Thus, the total error probability is:
P_{total} = n × (P_{header} + P_{data}) + (n – 1) × (P_{header} + Pdata) +...(P_{header} + P_{data}) = (n × (n + 1)/2) × ((P_{header} +P_{data})
One error impairs exactly one frame carrying one data block (Loss = 1). Thus, P_{total} is the loss of data blocks per cycle (Lpc = P_{total}).
Sum frame procedure
In the sum frame procedure, the complete frame is transmitted via all routes. The error probability thus is:
P_{total} = n × (P_{header} + n × P_{data})
Same as with individual frames, error probability
increases quadratically, but loss is much bigger. In case of impairment at various points in the line, the total loss comprises:
 Before the first node: all output data frames + all input data frames = 2n
 Before the second node: n – 1 output data frames + all input data frames = 2n – 1
 ...
 Before the last node: 1 output data frame + all input data frames = n + 1
 On the return path: all input data frames = n.,
The average amount of loss per transmission route is:
Loss = (5n + 1)/4 ≈ 1.25n
If P_{header} and P_{data} are approximately equal, the losses in the individual frame procedure (Loss_{i.f.}) and in the sum frame procedure (Loss_{s.f.}) are as follows:
P_{header} ≈ P_{data}
Loss_{i.f.}= n(n + 1) × P
Loss_{s.f.}= n(n + 1) × P × (1.25n)
The first approximation shows that, compared to the individual frame procedure, the sum frame procedure is more impaired by electromagnetic disturbances by a factor of 1.25n. In a line of 40 nodes, the number of EMC–related problems is 50 times as high in a sum frame system than in an individual frame system. This greatly restricts the practicality of the sum frame procedure and calls into question its suitability for applications under harsh EMC conditions. Figure 2 illustrates these aspects. We assumed a one in 100 million byte error probability and a payload of 40 bytes per node. Under these boundary conditions, there is a statistical loss of 0.5 bits per cycle in the individual frame procedure and 5 bits per cycle in the sum frame procedure. In other words, in case of massive interference, statistically there will be one error every half hour in a sum frame system but only one error every 24 hours in an individual frame system.
Fig. 2. Error probability (relative to the basic interference frequency) per number of nodes in the sum and individual frame procedures
Substantiation tests
First conclusions frompracticalmeasurements were:
 low sample tolerance – the results are reliable and highly reproducible;
 significant performance divergence in various Ethernet components...which suggests comparing various solutions;
 larger frames lead to a linear increase of transmission errors; this supports a crucial aspect of sum frame analysis; and
 there is a positive correlation between the number of frames and the error frequency.
The tests are being continued with the aim of developing a practical method for comparing the bus systems. Already, we can say that varieties of Ethernet are not necessarily the same and that users are welladvised to observe system component characteristics when deciding on the general network topology.
Sources
Jasperneite, Jürgen; Schumacher,Markus;Weber, Karl: Limits of Increasing the Performance of Industrial Ethernet Protocols. 12th IEEE Conference on Emerging Technologies and Factory Automation, Patras, Greece, Sep 2007.
