UDC 004.75 : 629.4.053 

Keyno M.Yu.

DATA PROCESSING for online monitoring system

Far Eastern state transport university

Работа посвящена направлениям развития систем мониторинга состояния локомотивов. Важной задачей является организация эффективной обработки массивов данных, собираемых бортовыми системами. Оперативный анализ данных позволяет своевременно выявлять отклонения в работе оборудования и предупреждать возникновение отказов.

The work is devoted to directions of development of monitoring the state of locomotives. An important task is to organize the efficient data processing for databases which are collected by the onboard systems. Operational data analysis will allow to detect deviations in the equipment and prevent the emergence of failures.

Locomotives, monitoring, data processing, reliability

Operational reliability is a key factor to reach new levels of productivity and safety of locomotives. Long time the manufacturers successfully trying to increase the reliability and achieve longer overhaul runs for diesels and electric locomotives. On Class I railroads in North America since early 80th, according FRA rules, are in operation the locomotives with 92-days inspection intervals. One of major players on market of diesel locomotives, General Electric since 2005 produces a new generation locomotives – the Evolution Series, with state of the art 184-days maintenance cycle. [1]. One of the key elements to ensure reliability of locomotive is a multilevel onboard digital control system.

In the paper [2] was compared reliability of old and new locomotives from Russia and South Africa, and shown, that for stabile locomotive operation on very long haul (3000 km and more) we need reach very low Failure Rate (lower than 8 fpmkm) and very high availability (more than 0.98). In current state, it is possible only if will be developed and implemented efficient online monitoring system.

Here is the second challenge – to develop multilevel distributed system which will be able doing non-stop measurement, will provide on board data pre-processing, gathering information from different onboard systems, transfer data to data processing center and process data from common database. Additionally, the final results of analysis must be available on web site or in the desktop and mobile user application. Such multi-loop system must work consistency for a long time and must have the own embedded mechanisms for fault state recognition and self-recovery. In this paper we will try describe common requirements to each level of system.

The lower level is onboard data acquisition (DAQ) subsystem. The main tasks for this layer are: measurement, signal conditioning, digital filtering, mean and RMS values calculation, data transfer. Engineers, who will develop software and hardware equipment for this layer must consider the big volume of channels, the different sample rates for different channels, and provide the simple algorithms for changing of configuration. The design of the onboard subsystem must be based on reliable and distributive hardware to minimize total cost of ownership. Whole details of building onboard DAQ subsystem are issues for a separate paper.

From data processing point of view, lower layer must provide effective filtering and fast calculation. Inappropriate technique, incorrect sequence or wrong parameters of filters may cause to significant signal distortion or cause artifacts in filtered signal. In common case for onboard system must be support next rule: if it possible, lower layer must send whole data with minimum filtering or correction to upper level. Only simple filtering must be used for screening of the obvious errors of measurements. For example, low level's filter for analog input (AI) or digital input (DI) channels must correct any missed single measure or short interference noise: all short-time changes, that can't present in real physical systems, must be ignored and replaced by previous data or by current calculated mean values. For counters low level filtering must prevent the emergence of unreal measurements, for example, if the wheel change the speed from 50 km/h to 380 km/h and back during 0,1 sec. For effective filtering low level digital processor must store enough data, what received in previous iterations. It will be better, if digital processor will support online adjusting for key parameters of filters and algorithm's. We are suggesting to implements bi-directional exchange protocol between low- and mid-level systems for automatically adjusting the parametric settings.

The calculations must be used for values, what can be calculated on upper level with use huge raw data only. For example, if we want to measure the frequency and phase of sinusoidal current (I) and voltage (U) signals with good precision we can use relatively simple algorithms on the low level and send to upper level only 4 bytes per cycle. But if we want doing calculation on the upper level, we must send two big arrays of 12 or 16-bit values per cycle. It cause 20, 100 or more times traffic in the field network without any real preferences.

Measured and calculated data from lower layer must be collected and pre-processed by onboard controller. Pre-processing is necessary when we are try to detect the failure or recognize the important state (incapability) by compare different parameters. This level can use sophisticated algorithms and need in strong timing synchronization. For example, when we watch for high voltage circuits (25 kV) on the electric locomotive, we can use only 5 sensors for recognize up to 17 important different state and deviations in pantograph, main switcher, traction transformers and protection circuits. Smart logic and inferences allow us to dynamically change the parameters of algorithms and, sometimes, the whole algorithm, depending on external conditions.

If will be detected the abnormal state, the structure of the output message can be changed for transmit information about failure or deviation. In the normal state onboard processor can send the short messages to upper level system relatively rare. But when alarm occurs, the volume and the frequency of messages must be increased. In the emergency cases, messages must be send on-by-one, without any delays and connection breaking.

The data, which collects on the upper level, are placing as linked structured tables in the database. Since the upper level system collect data from several objects, so all data must be analyzed in common time and space field. Many years before, the strong synchronizing in distributed systems had no easy solution. Now we are use the GPS signal not only for positioning, but for a time synchronizing first. GPS is at the present time the most competent system for time transfer. So, keeping all wired data in common database, we have unique ability to explore whole process of work for multiple locomotives. For instance, multivariate analysis can identify conditions, places and time where regular wheel spinning, system malfunction or traction equipment overheating occur. The main trouble there is detection of correspondence or correlation between discrete of short-time events and prolonged processes. In real conditions the one relatively fast event can invoke the long-time processes: driver drops the pressure in brake pipeline by main crane for 3 or 5 second, but brakes will be applied for minutes then. In our work we are tried to use the standard technique for data processing and found that only selected methods can be useful.

Finally, in multilevel online monitoring system, procedures of processing data must be divided by corresponding levels. Data processing realized by three relatively independent cycles, which linked only by data exchange protocols. Proposed system will solve the main task to raise the reliability of locomotives through building the efficient and very reliability monitoring system.

REFERENCES

1. GE Transportation. Rail Products. Evolution® Series Locomotive.

[http://www.getransportation.com/rail/rail-products/locomotives/evolutionr-series-locomotive.html]

2. M. Keyno, R.D. van der Meulen, Differences and similarities: learning from heavy haul in Cold and Heat // Proceedings 9th International Heavy Haul Conference «Railroading in Extreme Conditions», June 19 – 22, 2011, Calgary, Canada