Figure 1: P&H/CRCMining 2100BLE Research Shovel

Transcription

1 Introduction This paper presents the results of two projects worked on while studying as an occupational trainee with the CCMining Smart Machines Group in the Division of Mechanical Engineering at the University of Queensland. Both projects concern the P&H/CCMining BLE esearch Shovel, as shown in Figure : Figure : P&H/CCMining BLE esearch Shovel The shovel is property of P&H and is used for research only. The final goal of the research is an autonomously operating shovel, without intervention of an operator. To achieve this goal it is important to have a detailed and actual image of the surroundings of the shovel, i.e. to determine where to scoop and to avoid collisions with surrounding objects. The image is created with combined data from different sensors and computers. To ensure that the collected data describes the same moment in time, the different sensors and computers have to be time synchronized. This is the subject of the first project: Network time synchronization study. Precise time synchronization is essential in the fusion of range sensors and machine pose data from different computers, which are used to provide a detailed overview of the surroundings of the shovel. The Network Time Protocol (NTP) is a protocol which can be used for this kind of synchronization; however the performance of the synchronization is dependant of the used configuration. The performance of NTP running under a QNX operating system is tested in a laboratory environment and in the field on the P&H/CCMining BLE esearch Shovel. ecommendations are made about the best configuration to use.

2 The second project concerns the loading of the trucks which transport the product the shovel is digging for. To lengthen the lifetime of a truck it is important not to overload it, since otherwise i.e. the axels will wear out too fast. Therefore a system is installed on the shovel which is able to determine the amount of soil in the dipper, the so called Payload system. Measurements showed a drift in the output of this system, which is the subject of the second project: A study of thermal drift in the P&H loadpin instrumentation circuit. The so called Payload loadpin instrumentation circuit is used to determine the amount of soil scooped up with the dipper. A series of experiments is conducted to quantify the sensitivity of two key components in the loadpin circuit to variations in temperature. The results are correlated to theoretical predictions. For both projects separate reports are written, which are presented in the following.

3 Network time synchronization study Erik eichardt 8--8 Executive summary To determine the optimal settings for NTP time synchronization on the BLE shovel, a series of experiments are conducted to investigate the influence of a number of NTP server configuration options. The options included in the investigation relate to NTP client communication behaviour (iburst, burst), the polling interval (minpoll, maxpoll) and the number of servers (one, several). The experiments are conducted on computers both on the shovel and in the CC Mining Smart Machines laboratory and identified two possible settings for the best synchronization performance:. Iburst, synchronization to 4 servers, maxpoll 6 (the preferred setting).. Iburst, synchronization to 4 servers, minpoll 6. Synchronization with these settings results in the following performance: Converged mean 5 Minpoll Maxpoll 6.4 Converged standard Convergence time deviation [s] [min] The configuration file for the preferred setup is included in Appendix A. The following observations were made regarding the use of server options and NTP performance:. Synchronization is better (in terms of convergence time and converged offset variability) with four available time servers, rather than one, with the prefer command included for the highest stratum server.. Use of the burst function is not found to improve the time synchronization performance. 3. In general, the use of the maxpoll command to constrain the polling interval improves the converged mean and standard deviation. 4. Network latency (in the form of increased delay) correlates strongly with decreased time synchronization performance (increased offset). 3

4 Table of contents Introduction...5. Background...5. Experiment setup Performance measures...6 Synchronization using the iburst and burst server options...8. Experiment A...8 Aim...8 esults...8. Experiment B...3 Aim...3 esults Experiment C...8 Aim...8 esults Synchronization results using only the iburst server option Experiment A... Aim... esults Experiment B...6 Aim...6 esults Experiment C...3 Aim...3 esults Summary Conclusion...37 Appendix A

5 Introduction. Background The Network Time Protocol (NTP) is a protocol which is designed to synchronize computer clocks over a network, by exchanging packets including timestamps. This protocol is used to synchronize the computers on the shovel to the GPS time provided by the Applanix precision time source. There is a requirement for time synchronization to within ms of the provided GPS time to ensure position errors from scan data remain within an acceptable error budget. The aim of the experiments discussed in this report is to establish the performance of NTP with default configuration options, and improve this if possible. There are number of candidate options available for performance improvement, including:. The number of time servers. Synchronization to one or several time servers may influence the result of the synchronization.. The mode of communication with the server. Two options are considered. iburst speeds up the initial synchronization by sending a burst of eight packets instead of the usual one (this was adopted for all experiments). burst forces NTP to send a burst of eight packets to increase timekeeping quality with every data exchange with the server. 3. The polling interval. The minimum time interval between sending packets is defined using minpoll. The maximum time interval between sending the packets is defined by maxpoll. More complete definitions of the options (and the server command) can be found on and Experiment setup To determine the optimal settings for NTP time synchronization on the shovel, experiments on both the computers on the shovel and in the laboratory are conducted, varying different configuration parameters. The influence of the following three configuration parameters is investigated:. Synchronization to one or four NTP time servers.. Use of the burst command, in addition to iburst. 3. Use of the minpoll and maxpoll commands. The experiments are conducted on two computers in parallel (SAO and SAO ) on both the shovel and in the laboratory. The SAO computers are configured to 5

6 synchronize to several time servers, the SAO computers synchronize to one time server. The SAO computer in the laboratory is synchronized to four UQ time servers: ntp, ntp, ntp and ntp3. These time sources are all available within the UQ network. The ntp server is the highest stratum server, using a GPS reference clock. Accordingly, the prefer option for SAO is applied to the ntp server command. SAO synchronizes to ntp alone. The SAO computer on the shovel synchronizes to four time servers: the GUI on the shovel (providing the precision GPS time reference), ntp, ntp and a public time server. The GUI has the highest stratum, since it is synchronizing to GPS time from the Applanix. Accordingly, the prefer option from SAO is applied to the GUI server option. SAO synchronizes to the GUI alone. Two sets of experiments are conducted for each computer, on the basis of client communication behaviour:. Synchronization using both iburst and burst commands.. Synchronization using only the iburst command. Each of these experiments involves three sub-experiments, designed on the basis of polling interval options: A. No defined poll interval. B. Minpoll 6. C. Maxpoll 6. The results of the experiments are included in Sections and 3, and are summarized in Section 4..3 Performance measures The synchronization performance is quantified in terms of:. Convergence time: The time it takes the synchronization to achieve the offset requirement of ms.. Converged mean error: The mean time offset for the client computer after convergence. 3. Converged standard deviation in mean error: The standard deviation of the time offset for the client computer after convergence. With some configurations the ms offset requirement is not achieved. In those situations the convergence time is infinite. The mean error and standard deviation are calculated over the region where the synchronization performs best Note that ntp and ntp3 are only available within the UQ network. 6

7 (heuristically). In general, this follows a clearly evident transient period in the NTP client synchronization. In some cases the connection to the server is lost. The means and standard deviations are calculated over the time in which there is a connection to the server(s), following the transient behavior. For every experiment, figures are included to show the behavior of the synchronization. In the figures a division is made between coarse and fine convergence. Coarse convergence is characterized by gross steps in client time offset and large rate differences between the clock frequencies of the client and server. Fine convergence is characterized by a fine tuning of the clock frequency error between the client and server, without time offset step corrections. The offset is generally close to zero in the period of fine convergence. The division between coarse and fine convergence is generally clear from the data. 7

8 Synchronization using the iburst and burst server options. Experiment A Aim Investigate the performance of time synchronization on the laboratory and shovel computers using the iburst and burst commands, without defining a poll interval. esults Laboratory SAO : ntp offset :3 AM :38 AM :45 AM :5 AM 3: AM 3:7 AM 3:4 AM 3: AM 6 x -3 ntp delay 4 :3 AM :38 AM :45 AM :5 AM 3: AM 3:7 AM 3:4 AM 3: AM Figure : Coarse convergence, laboratory SAO, nopoll. x -3 ntp offset : PM : AM. ntp delay.5 : AM : PM : AM : PM Figure : Fine convergence, laboratory SAO, nopoll. 8

9 From Figure, it can be noted that the converged time synchronization is good. However, the offset diverges slightly when NTP automatically increases the polling interval at around 4am. The convergence time is approximately 5 hrs. Laboratory SAO ntp offset :3 AM :38 AM :45 AM :5 AM 3: AM 3:7 AM 3:4 AM 3: AM 4 x -3 ntp delay 3 :3 AM :38 AM :45 AM :5 AM 3: AM 3:7 AM 3:4 AM 3: AM Figure 3: Coarse convergence, laboratory SAO, nopoll. x -3 GPS offset 5-5 : PM : AM 8 x -3 GPS delay 6 4 : AM : PM : AM : PM Figure 4: Fine convergence, laboratory SAO, nopoll. It is noted that this synchronization does not meet the required offset performance of ms. 9

10 Shovel SAO.8 GPS offset :48 AM6: AM7: AM8:4 AM9:36 AM:48 AM: PM: PM:4 PM3:36 PM.5 x -3 GPS delay.5 4:48 AM6: AM7: AM8:4 AM9:36 AM:48 AM: PM: PM:4 PM3:36 PM Figure 5: Coarse convergence, shovel SAO, nopoll. 4 x -3 GPS offset - 9:36 AM:4 PM7: PM: AM4:48 AM9:36 AM:4 PM7: PM: AM4:48 AM9:36 AM 3 x -3 GPS delay 9:36 AM:4 PM7: PM: AM4:48 AM9:36 AM:4 PM7: PM: AM4:48 AM9:36 AM Figure 6: Fine convergence, shovel SAO, nopoll. These figures show that the settling time is very long, approximately 6 hours. However, after coarse convergence the performance steadily improves and the time synchronization meets our requirement. It should be noted that this profile of coarse convergence is unique among the results of Experiment A. The convergence time is approximately 43 hours. It should also be noted that the fourth (public) times server configured for this experiment was unable to be reached by the NTP client. This result, and those of subsequent SAO (shovel) experiments, are generated from three time sources (two external to the shovel), rather than four.

11 Shovel SAO GPS offset :4 AM 7:48 AM 7:55 AM 8: AM 8:9 AM 8:6 AM 8:4 AM 8:3 AM GPS delay - 7:4 AM 7:48 AM 7:55 AM 8: AM 8:9 AM 8:6 AM 8:4 AM 8:3 AM Figure 7: Coarse convergence, shovel SAO, nopoll.. GPS offset : AM : PM : AM : PM : AM : PM : AM 3 x -3 GPS delay : AM : PM : AM : PM : AM : PM : AM Figure 8: Fine convergence, shovel SAO, nopoll. The synchronization results from this experiment do not meet the required offset bounds of ms.

12 The results of Experiment A are summarized in Table. Table : Performance results of Experiment A. EXPEIMENT A Laboratory Shovel SAO SAO SAO SAO Converged mean [s] Converged standard deviation [s] Convergence time 5 43 [hrs]

13 . Experiment B Aim To investigate the time synchronization performance on the laboratory and shovel computers using the iburst and burst commands, with a defined minimum poll interval of 6, which corresponds to a minimum poll interval of 6 = 64seconds. Automatic increases of the poll interval are allowed. esults Laboratory SAO.8 ntp offset :48 AM 7: AM 9:36 AM : PM :4 PM 4:48 PM 7: PM. ntp delay.5 4:48 AM 7: AM 9:36 AM : PM :4 PM 4:48 PM 7: PM Figure 9: Coarse convergence, laboratory SAO, minpoll 6. x -3 ntp offset 5-5 : PM : AM : PM : AM : PM : AM.5 ntp delay..5 : PM : AM : PM : AM : PM : AM Figure : Fine convergence, laboratory SAO, minpoll 6. 3

14 It can be seen that the coarse convergence takes a very long time, approximately hours. The convergence time is much longer, 48 hours. The same coarse convergence behaviour is noted that was observed in the shovel SAO trial in Experiment A. Laboratory SAO.6 ntp offset : AM : PM : AM : PM : AM : PM : AM.5 ntp delay..5 : AM : PM : AM : PM : AM : PM : AM Figure : Synchronization result laboratory SAO, minpoll 6. We note that the performance of this synchronization does not meet the required offset performance of ms. Shovel SAO GPS offset :33 AM 7:4 AM 7:48 AM 7:55 AM 8: AM 8:9 AM 8:6 AM 8:4 AM 8:3 AM.5 x -3 GPS delay.5 7:33 AM 7:4 AM 7:48 AM 7:55 AM 8: AM 8:9 AM 8:6 AM 8:4 AM 8:3 AM Figure : Coarse convergence, shovel SAO, minpoll 6. 4

15 x -3 GPS offset - -4 : AM : PM : AM : PM : AM : PM : AM.5 x -3 GPS delay.5 : AM : PM : AM : PM : AM : PM : AM Figure 3: Fine convergence, shovel SAO, minpoll 6. Figure shows that the synchronization results from this experiment are reasonable, but do not meet the required specifications. The offset fluctuates outside ± ms. The various levels that are observed in the converged time offset in Figure 3 are the result of conflicting offsets returned by sequential packets (the result of the burst option). The packets with increased offset also have increased delay. We attribute the different levels of offset to changes in the delay resulting from (presumably) different routing of packets from client to server. 5

16 Shovel SAO.5 GPS offset :4 PM :3 PM :38 PM :45 PM :5 PM : AM :7 AM :4 AM.5 x -3 GPS delay.5 :4 PM :3 PM :38 PM :45 PM :5 PM : AM :7 AM :4 AM Figure 4: Coarse convergence, shovel SAO, minpoll 6.. GPS offset. -. : AM4:48 AM9:36 AM:4 PM7: PM: AM4:48 AM9:36 AM:4 PM7: PM: AM 3 x -3 GPS delay : AM4:48 AM9:36 AM:4 PM7: PM: AM4:48 AM9:36 AM:4 PM7: PM: AM Figure 5: Fine convergence, shovel SAO, minpoll 6. Figures 4 and 5 show that the settling time is short, however the performance does not meet the specified requirement. 6

17 The results of Experiment B are summarized in Table. Table : Performance results of Experiment B. EXPEIMENT B Laboratory Shovel SAO SAO SAO SAO Converged mean [s] Converged standard deviation [s] Convergence time 48 4 [hrs] 7

18 .3 Experiment C Aim To investigate the time synchronization performance on the laboratory and shovel computers using the iburst and burst commands, with a defined maximum poll interval of 6, which corresponds to a maximum poll interval of 6 = 64seconds. The poll interval cannot automatically increase with this setting. esults Laboratory SAO.3 ntp offset.. -. :4 PM:3 PM:38 PM:45 PM:5 PM: AM:7 AM:4 AM: AM:8 AM:36 AM 3 x -3 ntp delay :4 PM:3 PM:38 PM:45 PM:5 PM: AM:7 AM:4 AM: AM:8 AM:36 AM Figure 6: Coarse convergence, laboratory SAO, maxpoll 6. 6 x -3 ntp offset 4 - : AM:4 AM4:48 AM7: AM9:36 AM: PM:4 PM4:48 PM7: PM9:36 PM: AM. ntp delay.5 : AM:4 AM4:48 AM7: AM9:36 AM: PM:4 PM4:48 PM7: PM9:36 PM: AM Figure 7: Fine convergence, laboratory SAO, maxpoll 6. 8

19 Figure 7 shows that the performance is generally within the required specifications. It is noted that some observations lie outside the +/-ms window. Since the burst command is employed, these individual points are one of a set of eight packets. The other points of the burst group lie within the specified tolerance. The packets with increased offsets are observed to have increased delay, suggesting that network variability is a possible cause of the variable offsets. The convergence time is approximately 4 hours. Laboratory SAO.6 GPS offset : PM: AM 4:48 AM 9:36 AM :4 PM 7: PM: AM 4:48 AM 9:36 AM.5 GPS delay..5 7: PM: AM 4:48 AM 9:36 AM :4 PM 7: PM: AM 4:48 AM 9:36 AM Figure 8: Synchronization result laboratory SAO, maxpoll 6. Figure 8 shows that the connection to the time server is lost when the computer is trying to synchronize to the single time server. When there is a connection the synchronization is very good. Performance results are calculated over the last hours where a connection is established. 9

20 Shovel SAO.8 GPS offset.6.4. : AM :4 AM 4:48 AM 7: AM 9:36 AM : PM :4 PM 3 x -3 GPS delay : AM :4 AM 4:48 AM 7: AM 9:36 AM : PM :4 PM Figure 9: Coarse convergence, shovel SAO, maxpoll 6. 3 x -3 GPS offset - : PM :4 PM 4:48 PM 7: PM 9:36 PM : AM :4 AM 4:48 AM 3 x -3 GPS delay : PM :4 PM 4:48 PM 7: PM 9:36 PM : AM :4 AM 4:48 AM Figure : Fine convergence, shovel SAO, maxpoll 6. With this configuration the convergence time is about 7 hours.

21 Shovel SAO. GPS offset : AM : AM : AM : AM : AM : AM : AM.3 GPS delay.. : AM : AM : AM : AM : AM : AM : AM Figure : Synchronization result shovel SAO, maxpoll 6 From Figure, it is noted that the performance of this synchronization does not meet the specified requirement of ms. The results of Experiment C are summarized in Table 3. Table 3: Performance results of Experiment C EXPEIMENT C Laboratory Shovel SAO SAO * SAO SAO 5 Converged mean [s] Converged standard deviation [s] Convergence time 4 7 [hrs] * Data from limited set 5

22 3 Synchronization results using only the iburst server option 3. Experiment A Aim To investigate the time synchronization performance on the laboratory and shovel computers using only the iburst command, without defining a poll interval. esults Laboratory SAO.3 ntp offset :9 AM4:33 AM4:48 AM5: AM5:6 AM5:3 AM5:45 AM6: AM6:4 AM6:8 AM.6 ntp delay.4. 4:9 AM4:33 AM4:48 AM5: AM5:6 AM5:3 AM5:45 AM6: AM6:4 AM6:8 AM Figure : Coarse convergence, laboratory SAO, nopoll..3 ntp offset :48 AM9:36 AM:4 PM7: PM: AM4:48 AM9:36 AM:4 PM7: PM: AM4:48 AM.6 ntp delay.4. 4:48 AM9:36 AM:4 PM7: PM: AM4:48 AM9:36 AM:4 PM7: PM: AM4:48 AM Figure 3: Fine convergence, laboratory SAO, nopoll.

23 Laboratory SAO ntp offset : AM4:48 AM9:36 AM:4 PM7: PM: AM4:48 AM9:36 AM:4 PM7: PM: AM.6 ntp delay.4. : AM4:48 AM9:36 AM:4 PM7: PM: AM4:48 AM9:36 AM:4 PM7: PM: AM Figure 4: Synchronization result, laboratory SAO, nopoll. Figure 4 shows that the connection between the client and the server was lost for a large period. The performance results are calculated over the first set of data until connection is lost for the first time. Shovel SAO GPS offset :57 AM 4:4 AM 4: AM 4:9 AM 4:6 AM 4:33 AM 4:4 AM 4:48 AM 6 x -4 GPS delay 4 3:57 AM 4:4 AM 4: AM 4:9 AM 4:6 AM 4:33 AM 4:4 AM 4:48 AM Figure 5: Coarse convergence, shovel SAO, nopoll. 3

24 x -3 GPS offset : AM4:48 AM9:36 AM:4 PM7: PM: AM4:48 AM9:36 AM:4 PM7: PM: AM.5 x -3 GPS delay.5 : AM4:48 AM9:36 AM:4 PM7: PM: AM4:48 AM9:36 AM:4 PM7: PM: AM Figure 6: Fine convergence, shovel SAO, nopoll. Shovel SAO. GPS offset. -. : AM : PM : AM : PM : AM : PM.5 x -3 GPS delay.5 : AM : PM : AM : PM : AM : PM Figure 7: Synchronization result, shovel SAO, nopoll. Figure 7 shows that the performance of this synchronization does not meet the required offset performance requirement of ms. 4

25 The results of experiment A are summarized in Table 4. Table 4: Performance results of Experiment A. EXPEIMENT A Laboratory Shovel SAO SAO * SAO SAO Converged mean [s] Converged standard deviation [s] Convergence time [hrs] * Data from limited set 5

26 3. Experiment B Aim To investigate the time synchronization performance on the laboratory and shovel computers using only the iburst command, with a defined minimum poll interval of 6, which corresponds with a minimum poll interval of 6 = 64seconds. Automatic increases of the poll interval are allowed. esults Laboratory SAO.5 ntp offset : AM8:4 AM9:36 AM:48 AM: PM: PM:4 PM3:36 PM4:48 PM6: PM7: PM. ntp delay.5 7: AM8:4 AM9:36 AM:48 AM: PM: PM:4 PM3:36 PM4:48 PM6: PM7: PM Figure 8: Coarse convergence, laboratory SAO, minpoll 6. 6 x -3 ntp offset 4 - : PM : AM : PM : AM : PM : AM. ntp delay.5 : PM : AM : PM : AM : PM : AM Figure 9: Fine convergence, laboratory SAO, minpoll 6. 6

27 The convergence time for this synchronization is approximately 54 hours. Laboratory SAO.5 ntp offset.5 : AM: AM:4 AM3:36 AM4:48 AM6: AM7: AM8:4 AM9:36 AM:48 AM 6 x -3 ntp delay 4 : AM: AM:4 AM3:36 AM4:48 AM6: AM7: AM8:4 AM9:36 AM:48 AM Figure 3: Coarse convergence, laboratory SAO, minpoll 6. 6 x -3 ntp offset 4 - : AM : PM : AM : PM : AM : PM : AM. ntp delay.5 : AM : PM : AM : PM : AM : PM : AM Figure 3: Fine convergence, laboratory SAO, minpoll 6. Figure 3 shows that during the coarse convergence time connection is lost a few times. Figure 3 shows that after convergence the performance improves. However the offset performance requirements are not met. 7

28 Shovel SAO GPS offset :36 AM 3:43 AM 3:5 AM 3:57 AM 4:4 AM 4: AM 4:9 AM 4:6 AM 6 x -4 GPS delay 4 3:36 AM 3:43 AM 3:5 AM 3:57 AM 4:4 AM 4: AM 4:9 AM 4:6 AM Figure 3: Coarse convergence, shovel SAO, minpoll 6. x -3 GPS offset - : AM : AM : AM : AM : AM x -3 GPS delay 5-5 : AM : AM : AM : AM : AM : AM Figure 33: Fine convergence, shovel SAO, minpoll 6. The synchronization using this configuration performs within requirements. However, it should be noted that when time increases, delay also increases resulting in a larger spreading of the offset. This could become a problem on larger time scales. The convergence time is about 35 minutes. 8

29 Shovel SAO. GPS offset : AM : PM : AM : PM : AM : PM : AM : PM 3 x -3 GPS delay : AM : PM : AM : PM : AM : PM : AM : PM Figure 34: Synchronization result, shovel SAO, minpoll 6. Figure 34 shows that this configuration is not appropriate for the synchronization, since the requirements are not met. The results of Experiment B are summarized in Table 5. Table 5: Performance results of Experiment B EXPEIMENT B Laboratory Shovel SAO SAO * SAO SAO Converged mean [s] Converged standard deviation [s] Convergence time 54.6 [hrs] * Data from limited set 9

30 3.3 Experiment C Aim To investigate the time synchronization performance on the laboratory and shovel computers using only the iburst command, with a defined maximum poll interval of 6, which corresponds to a maximum poll interval of 6 = 64seconds. The poll interval cannot automatically increase with this setting. esults Laboratory SAO.5 ntp offset :36 PM: AM:4 AM4:48 AM7: AM9:36 AM: PM:4 PM4:48 PM7: PM9:36 PM. ntp delay.5 9:36 PM: AM:4 AM4:48 AM7: AM9:36 AM: PM:4 PM4:48 PM7: PM9:36 PM Figure 35: Coarse convergence, laboratory SAO, maxpoll 6. 6 x -3 ntp offset 4-4:48 PM6: PM7: PM8:4 PM9:36 PM:48 PM: AM: AM:4 AM3:36 AM4:48 AM. ntp delay.5 4:48 PM6: PM7: PM8:4 PM9:36 PM:48 PM: AM: AM:4 AM3:36 AM4:48 AM Figure 36: Fine convergence, laboratory SAO, maxpoll 6. 3

31 The convergence time for this synchronization is approximately hours. Laboratory SAO.3 ntp offset : PM: AM 4:48 AM 9:36 AM :4 PM 7: PM: AM 4:48 AM 9:36 AM. ntp delay.5 7: PM: AM 4:48 AM 9:36 AM :4 PM 7: PM: AM 4:48 AM 9:36 AM Figure 37: Synchronization result, laboratory SAO, maxpoll 6. Figure 37 shows that by using this configuration, the connection to the time server is lost. However, when there is a connection the synchronization is very good, as can be seen from Figure 38. Performance statistics are calculated based on the behavior shown in Figure 38. Again, network latency (increased delay) correlates strongly with observations of increased offset. The author notes that it is a regular occurrence that the connection to ntp is lost, specifically when this is the only server used by the client. It is suspected that the server itself is responsible, possible preventing misuse by an errant client (not sharing the NTP load ). No concrete evidence of this was observed. 3

32 5 x -3 ntp offset -5 - : PM:4 PM3:36 PM4:48 PM6: PM7: PM8:4 PM9:36 PM:48 PM: AM. ntp delay.5 : PM:4 PM3:36 PM4:48 PM6: PM7: PM8:4 PM9:36 PM:48 PM: AM Figure 38: Synchronization performance when connection is established. 3

33 Shovel SAO GPS offset : AM 6:7 AM 6:4 AM 6: AM 6:8 AM 6:36 AM 6:43 AM 6:5 AM 6 x -4 GPS delay 4 6: AM 6:7 AM 6:4 AM 6: AM 6:8 AM 6:36 AM 6:43 AM 6:5 AM Figure 39: Coarse convergence, shovel SAO, maxpoll 6. x -3 GPS offset - : PM : AM : PM : AM : PM : AM x -3 GPS delay 5-5 : PM : AM : PM : AM : PM : AM Figure 4: Fine convergence, shovel SAO, maxpoll 6. The synchronization using this configuration performs within requirements, however it should be noted that when time increases, delay also increases resulting in a larger spreading of the offset. This could become a problem on larger time scales. The convergence time is approximately 35 minutes. 33

34 Shovel SAO.5 GPS offset : AM : PM : AM : PM : AM : PM : AM 3 x -3 GPS delay : AM : PM : AM : PM : AM : PM : AM Figure 4: Synchronization result, shovel SAO, maxpoll 6. Figure 4 shows that this configuration is not appropriate for the synchronization, since the offset requirement of ms is not met. The results of Experiment C are summarized in Table 6. Table 6 Performance results of Experiment C. EXPEIMENT C Laboratory Shovel SAO SAO * SAO SAO Converged mean [s] Converged standard deviation [s] Convergence time.6 [hrs] * Data from limited set 34

35 4 Summary The results from all experiments are summarized in Tables 7-. Table 7: esults of laboratory computer SAO synchronizing to four NTP time servers. esults of synchronization to UQ time server: NTP SPECIFIED OPTIONS iburst, burst Iburst Converged mean [s] Nopoll 3. Converged standard deviation [s] Minpoll Maxpoll Nopoll Minpoll Maxpoll Convergence time [hrs] Table 8: esults of laboratory computer SAO synchronizing to one NTP time server. esults of synchronization to UQ time server: NTP SPECIFIED OPTIONS iburst, burst Iburst Converged mean [s] Converged standard deviation [s] 5 Nopoll 5..7 Minpoll 6..5 Maxpoll 6* Nopoll* -.3. Minpoll 6*.3. 5 Maxpoll 6* Convergence time [hrs] Table 9: esults of shovel computer SAO synchronizing to three NTP time servers. esults of synchronization to GPS time server: SPECIFIED OPTIONS iburst, burst Converged mean [s] Nopoll 4.7 Minpoll 6.95 Converged standard deviation [s] Convergence time [hrs] 35

36 Iburst Maxpoll Nopoll -.. Minpoll Maxpoll Table : esults of shovel computer SAO synchronizing to shovel GPS time server alone. esults of synchronization to GPS time server: SPECIFIED OPTIONS iburst, burst Iburst Converged mean [s] Converged standard deviation [s] Nopoll Minpoll Maxpoll Nopoll.6.37 Minpoll Maxpoll Convergence time [hrs] 36

37 5 Conclusion In general, four conclusions can be made regarding the server options and NTP performance:. Synchronization is better with four available time servers, rather than one, with the prefer command included for the highest stratum server. It should be noted that the fourth server defined for the shovel SAO experiments was not reached during the experiments.. Use of the burst function is not found to improve the time synchronization performance. 3. In general, the use of the maxpoll command to constrain the polling interval improves the converged mean and standard deviation. 4. Network latency (in the form of increased delay) correlates strongly with decreased time synchronization performance (increased offset). According to the results from the experiments, there are two possible settings for time synchronization on the shovel, resulting in the best performance:. Iburst, synchronization to 4servers, maxpoll 6 (the preferred setting).. Iburst, synchronization to 4servers, minpoll 6. These settings both result in a performance which meets the requirements. However, it should be noted that the synchronization offset increases with increasing delay over the duration of the experiment. Since the delay continues increasing with increasing time, this could become a problem on large time scales. It is recommended to conduct an extended trial of the proposed NTP settings to determine the long-term performance. The configuration file for the recommended NTP setup is included in Appendix A. 37

38 Appendix A # Example for the shovel client setup server prefer iburst maxpoll 6 server server # Define place to save driftfile driftfile /etc/ntp/ntp.drift # Define place and type of synchronization statistics to save statsdir /etc/ntp/ntpstats/ statistics loopstats peerstats filegen loopstats file loopstats type pid enable filegen peerstats file peerstats type pid enable 38

39 A study of thermal drift in the P&H loadpin instrumentation circuit. Erik eichardt, Anthony eid, oss McAree, CCMining Wednesday, November 5, 8 Executive Summary This report explores the sources of the observed drift in the measurement signals from the Payload loadpin across several field installations. This drift appears to be correlated with temperature. The aim of this report is to investigate the thermal stability of two components in this circuit, specifically: The Dataforth signal conditioning unit which amplifies bridge voltages to a 4- ma signal for transmission down the length of the boom to the payload computer; and The loadpin cable connecting the loadpin to the Dataforth signal conditioning unit. The methodology of the study is to conduct a series of experiments that aim towards establishing the thermal stability of these components. The experiments, and their aims are: Experiment A: Establish Effect of temperature on the bridge excitation voltage from the Dataforth. Experiment B: Effect of temperature on the gain of the Dataforth signal conditioner. Experiment C: Effect of changes in the resistance of bridge excitation circuit as a function of temperature. Experiment D: Influence of loadpin cable resistance change on signal measured at payload box. The findings of the study are: Experiment A: A.99 μv/v deviation in bridge excitation voltage is observed across the temperature range -5 o C. When the Dataforth is cold, some initial drift in the excitation voltage is observed, however this settles within seconds applying power. The Dataforth does not meet its excitation voltage specification. Experiment B An equivalent variation of 5 μv/v in bridge voltage is observed over a temperatures from to 5 degrees when a 4mV reference signal is applied. 39

40 An equivalent variation of 7.86 μv/v in bridge voltage is observed over this temperature range when the input is V (that is, short circuited). Experiment C Temperature changes in the AWG cable connecting the Dataforth to the loadpin full bridge influence the excitation voltage, resulting in a bridge output deviation of.5 μv/v. Experiment D With a 4mV precision voltage source, the equivalent bridge voltage varies by.4 μv/v. When the bridge is short circuited, voltage deviations of.7 μv/v are observed. The changes in voltage are not correlated with the temperature of the AWG cable. The main conclusion from the study is that the drift due to thermal loading for these conditions is individually and collectively within a μv/v error budget. 4

41 Table of contents Introduction...4 An analysis of the expected change in bridge voltage due to changes in resistance of the loadpin cable due to thermal loading Experimental results Experiment A: Effect of temperature on the bridge excitation voltage from the Dataforth...47 Aim...47 Methodology...47 esults...48 Conclusion Experiment B: Effect of temperature on the gain of the Dataforth signal conditioner...5 Aim...5 Methodology...5 esults...5 Conclusions Experiment C: Effect of changes in the resistance of bridge excitation circuit as a function of temperature...56 Aim...56 Methodology...56 esults...57 Conclusion Experiment D: Influence of loadpin cable resistance change on signal measured at payload box...59 Aim...59 Methodology...59 esults...6 Conclusion Conclusion...63 Appendix A: Experimental Protocols...64 Experiment A...64 Experiment B...65 Experiment C...66 Experiment D...67 Appendix B: Oscilloscope settings

42 Introduction Figure shows the layout of the loadpin instrumentation circuit. The major components are: Loadpin bridge a full Wheatstone bridge embedded in the boom-point sheave pin. The loadpin cable This connects the loading to a signal conditioning amplifier. The cable specification is a 7.6 m (5ft) AWG cable. A Dataforth DSCA38-9C strain-gauge bridge signal conditioner. This unit converts the bridge voltage to a 4-mA signal. Boom cable carries the 4- ma signal down the boom to the payload computer. Load resistors 5Ohm resistor used to convert the 4-mA signal to a voltage that can be read by an analog-to-digital converter. STX4 the analog to digital converter used to read loadpin signal. Figure : Loadpin data collection setup. This report presents an experimental investigation of the thermal stability of two components in the strain-bridge instrumentation, namely the: Dataforth signal conditioner, and AWG cable. Along with the loadpin itself, these are the components that are expected to see significant thermal loading. 4

43 Four experiments are conducted, with the following aims: Experiment A: Establish Effect of temperature on the bridge excitation voltage from the Dataforth. Experiment B: Effect of temperature on the gain of the Dataforth signal conditioner. Experiment C: Effect of changes in the resistance of bridge excitation circuit as a function of temperature. Experiment D: Influence of loadpin cable resistance change on signal measured at payload box. Experiment C is also investigated by theoretical analysis. The structure of the report is as follows. Section presents a theoretical analysis of the expected effect of the resistance changes in the loadpin cable due to temperature variation. Section 3 presents results from experimental studies. The experimental protocols are given as Appendix A. A discussion of oscilloscope resolution is presented in Appendix B. 43

44 An analysis of the expected change in bridge voltage due to changes in resistance of the loadpin cable due to thermal loading Figure shows the electrical circuit used in this analysis. The Wheatstone bridge comprised the four resistors labeled, 3, 4, and 5 which have been selected to give a bridge imbalance equivalent to mv for a bridge excitation of V. The loadpin cable resistance is represented by. The objective of the analysis is to understand how changes in influence the voltage measured across points BD, namely the voltage Vg under the assumption that no current flows through Vg. (In practice the voltage Vg is an input to the Dataforth unit; the input impedance of the Dataforth was measured to be M Ohm. ) Figure : Electrical circuit representing loadpin bridge. The loadpin cable core is made of copper, with a diameter of.64 mm. With the 8 known resistively of copper (.7 Ω m) at degrees C and the fact that the cable is 7.6 m long, the resistance of the cable at room temperature can be calculated by using: ρl = =. Ω A With the known temperature coefficient (α) of copper (.39 Ω/K) the following relation for the temperature dependence of the resistance can be derived: = + [ + α ( T )] T Now the relationship between the excitation voltage (the input voltage of the loadpin full bridge), and the bridge output voltage V g can be derived by making use of Kirchhoff s current and voltage laws: 44

45 )] ( [ ) ( ) ( ) ( I V V V I V I I I V V V OA exc AB ABC ABC AB ADC ABC OA axc ADC ABC + = + = = + = + = = Using the same relations V AD can be derived, which results in: )] ( [ I V V OA exc AD + = Assuming there is no current flowing between points B and D through V g, the bridge output voltage is the difference between V AB and V AD and is given by the following: )] ( [ ) )( ( I V V OA exc g + + = Now by expressing the current I OA in terms of the excitation voltage and the resistors, this expression can be substituted into the derived formula for V g which will result in the desired relationship: ( ) ( ) ( ) ( ) = = = + = ) )( ( ) ( ) )( ( ) )( ( ) )( ( )] ( [ V V V V I I V I I I s s g s OA OA s ADC ABC OA This linear expression for V g can be implemented in Matlab to determine the change in bridge output voltage due to changing. The nominal resistances of the bridge resistors (,3,4,5 ) are 35 Ω. They increase in resistance when a load is applied. The voltage output of the bridge is bounded between and mv; to be well inside this range, the values for the resistances are chosen such that the bridge output is mv. By varying the temperature, the value for will change causing a change in the bridge output voltage. This physics has been simulated in Matlab using the script to given in Appendix A. Figure 3 shows the change in bridge output voltage as a function of temperature change from a nominal room temperature ( degrees). For example a dt of -

46 degrees C implies an environmental temperature of degrees. The bridge excitation voltage in this simulation is V consistent with the actual system. Figure 3: Bridge voltage drop as function of temperature This figure shows that with increasing ambient temperature, the bridge voltage drops. However this change is in the sub-micro volt range per volt of bridge excitation. The overall voltage drop for a temperature difference of 65 degrees is only.447 μv /V. In theory changes in the load pin cable do not account for observed drift in load pin signals. This conclusion will be experimentally tested by Experiment C. 46

47 3 Experimental results In the following of this report, four experiments are conducted with for each experiment separate results and conclusions. 3. Experiment A: Effect of temperature on the bridge excitation voltage from the Dataforth. Aim To establish if the bridge excitation voltage produced by the Dataforth strain gauge amplifier varies with temperature. Bridge output voltages are proportional to the bridge excitation voltage. Methodology The setup used to address this aim is shown in Figure 4. Figure 4: Excitation voltage test. Here, PS is a power supply used to power the Dataforth strain gauge amplifier; DF is the Dataforth. The block labeled OS is an oscilloscope. A resistor on the OS side of the diagram acts as a proxy for the strain gauge bridge; the resistor on the PS side of the diagram is the load resistor across the 4-mA loop. Both are precision 35 Ohm resistors. The excitation voltage is measured with an oscilloscope. The protocol for the experiment is given in Appendix A. The Dataforth unit is placed in an isolated environment, which will vary between three ambient temperatures: degrees (Dataforth placed in an ice bath); 3 degrees (Dataforth at room temperature); and 5 degrees (Dataforth heated by lamp). Before starting the experiment, the Dataforth is heated or cooled until the desired temperature is reached. The remaining equipment is installed such that the experiment can be started as soon as the Dataforth has reached the desired temperature. Each experiment will last 5 seconds, in which the output voltage is measured and logged with an oscilloscope. Every set of measurements is repeated three times. 47

48 esults Table gives the measurement results with uncertainties for the different temperatures. The data is graphically represented in Figure 5. Table : Average excitation voltages for different Dataforth temperatures Experiment Measurement Dataforth temperature ( o C) Excitation voltage (V) degrees 4 ±.5.7 ±. ±.5.8 ±. 3 8 ±.5.86 ±. 3 degrees 5 degrees 6 ±.5.87 ±. 7 ±.5.89 ±. 3 9 ±.5.9 ±. 5 ±.5.98 ±. 5 ±.5.99 ±. 3 5 ±.5. ±. Figure 5: Experiment A measurement results. The range of deviation between the load resistor measurements is 7 mv. Note this includes the first measurement of.7 volts, which seems to be an outlier. 48

49 Therefore this value is neglected and the second largest difference is calculated to be mv. To express the temperature dependent voltage deviations in a more general way, this change in excitation voltage is translated to a deviation in the bridge output voltage (in μv/v). A change of mv in the excitation voltage results in a deviation of.99 μv/v in the bridge output voltage. Figure 6 shows the voltages converted to equivalent bridge voltages, using the nominal bridge imbalance of mv/v. Figure 6: Experiment A measurement results as equivalent bridge voltages. A nominal bridge output of mv/v is assumed. Other observations Quality of power regulation During the conduct of these experiments a transient (ringing) was observed on the excitation voltage. The transient induces a peak-to-peak voltage of.v on the V signal. The transient repeats at approximately 75kHz. Figure 7 shows the dynamic character of the transient. Figure 8 the frequency of repetition. Several Dataforth units tested showed the same characteristic. Similar transients can be measured at several parts of the circuit including across the load resistor. The anti-aliasing filter used on installed systems would attenuate 49

50 changes due to this transient. However, it is noted that such a significant transient is undesirable from an instrumentation amplifier. There is a strong case for reviewing the selection of this device. Figure 7: Transient on the excitation voltage. Figure 8: epetition of transients on the excitation voltage. Voltage drift on application of power to the Dataforth When the Dataforth is connected to the power source, the excitation voltage rises to a (approximately) constant value. This is shown in Figure 9. Note the spikes are the aliased transients referred to in the previous section and are not of interest here. 5

51 The cause is thought to be warming of the circuit. When the Dataforth is initially connected to the power source, the electronics are cold; when power is applied the temperature of circuit components increases correlated with an increase in the excitation voltage. Figure 9: Initial increasing excitation voltage when Dataforth is cooled. Dataforth excitation voltage is out of specification The Dataforth specification states the excitation voltage will be.v +/-.3V. The average measured voltage is.v. This can be seen in Figures 6, 7 and 9. Several units have been tested an found to be similarly out of specification. Conclusion In summary: A.99 μv/v deviation in bridge excitation voltage is observed across the temperature range -5 o C. When the Dataforth is cold, some initial drift in the excitation voltage is observed, however this settles within seconds. The Dataforth does not meet its excitation voltage specification. 5

52 3. Experiment B: Effect of temperature on the gain of the Dataforth signal conditioner. Aim To establish the level of variation of in the gain of the Dataforth as a function of temperature. Methodology To Dataforth amplifies the bridge voltage to a 4-mA regulated current. Figure shows the experimental layout used to establish the effect of temperature changes on this gain. Figure : Dataforth gain experiment. The Dataforth unit gets its power from the power source, and receives a voltage from the voltage source which simulates a measured load on the loadpin bridge. The Dataforth transforms this signal into a current, which is measured by the oscilloscope over a 35 Ω prevision resistor. The Dataforth is placed into the same isolated environment as in Experiment A, and the temperature is varied between the, 3 and 5 degrees. The output current is measured and logged with the oscilloscope for 5 seconds. This procedure is also repeated three times. A variation of this experiment is obtained by setting the voltage source to V (implemented by a short circuit). Here the Dataforth should give a constant output of 4 ma. esults The input voltage which simulates a constant load over the loadpin full bridge is provided by the Multi-function calibrator 7 stable voltage source. This voltage source is designed to provide a stable voltage output, however to calibrate the output some tests are performed measuring the output voltage with the Tektronix TDS 4B oscilloscope. The FLUKE multimeter is first used to calibrate the output voltage to 4 mv, after that the oscilloscope is connected to the voltage source to monitor the output. Input from a regulated voltage source Table contains the measurement results with uncertainties for the different temperatures of the Dataforth, which are also graphically represented in Figure in terms on the equivalent loadpin bridge voltage. 5

53 Table : Average Dataforth output voltages for different Dataforth temperatures Experiment Measurement Dataforth Load resistor Equivalent bridge temperature voltage (V) ( o C) output voltage [mv] degrees 8 ± ± ± ± ± ± degrees 5 degrees 7 ± ± ± ± ± ± ± ± ± ± ± ± The collected data show that the transformation gain from voltage to current is affected by temperature changes. The largest measurement difference in Dataforth output voltage is 3 mv, on a scale of.4-7 V. From the measured average of about 5.3 V, this is a deviation of.5%. To translate the measured Dataforth output voltages to the bridge voltage the following relationship is used: V Vb =

54 Figure : Experiment measurement results as equivalent bridge voltage. The derived results (Table ) show a deviation of maximum.5 mv in the bridge output voltage. This corresponds to a change of 5 μv/v in the bridge output voltage. Short circuit input When the Dataforth signal conditioner has no input voltage from the bridge (simulated by a short circuit), it should provide a constant output current of 4 ma. Measured over a 35 Ω resistance, this coheres with a constant output voltage of.4 V. The results of the Dataforth output voltage measurement with the short circuited bridge input channel are shown in Table 3. 54

55 Table 3: Average Dataforth output voltages for zero bridge input and different Dataforth temperatures. Experiment Measurement Dataforth Load resistor Equivalent temperature voltage bridge output ( o C) (V) voltage [mv] degrees 4 ± ± ± ± ±.5.44 ±..5 3 degrees 5 degrees 8 ± ±..4 3 ±.5.43 ± ±.5.43 ±.. 6 ±.5.45 ± ±.5.45 ± ±.5.44 ±..86 To translate the measured output voltages from the Dataforth unit to the bridge voltages, the same relationship is used as described in the previous experiment. The results are depicted graphically in Figure. Figure : Experiment B short-circuited voltage source results, as equivalent bridge voltage. The results show a maximum deviation in bridge output voltage of.786 mv. This is equivalent to a change of 7.86 μv/v in the bridge output voltage. 55

56 Conclusions In summary: An equivalent variation of 5 μv/v in bridge voltage is observed over temperatures from to 5 degrees when a 4mV reference signal is applied. An equivalent variation of 7.86 μv/v in bridge voltage is observed over the - 5 temperature range when the input is V (that is, short circuited). 3.3 Experiment C: Effect of changes in the resistance of bridge excitation circuit as a function of temperature. Aim To establish the level of observed due to changes in the resistance of the bridge excitation circuit. This experiment looks to validate the analytical result presented in Section which showed that changes in the resistance of the loadpin cable have a negligible effect. Methodology The AWG cable has a core which is made of copper. It is known that copper has a temperature dependant resistance; therefore experiments are done under different circumstances with the same 5 ft long cable as used in the loadpin setup. To determine the effect of change in resistance of the cable, the arrangement in Figure 3 is used. Figure 3: Testing the effect of loadpin cable resistance changes on bridge excitation. In this experiment the Dataforth is powered by the power source, and the excitation voltage outputted by the Dataforth and going through the AWG cable is measured by the oscilloscope, using a 35 Ω resistance. The cable is placed in the same isolated environment as in the previous experiments, and the temperature is changed from to 3 to 5 degrees. For every temperature the experiment is conducted, and the data is logged for 5 seconds with the oscilloscope. Every set of measurements is repeated for three times. 56

57 esults The results of the excitation voltage measured over the AWG cable for different temperatures are shown in Table 4, which are also graphically represented in Figure 4. Table 4: Average excitation voltages for different AWG cable temperatures. Experiment Measurement Cable temperature ( o C) Excitation voltage (V) degrees.75 ±..77 ± ±. 3 degrees 5 degrees 6 ±.5.74 ±. 4 ±.5.79 ±. 3 5 ±.5.76 ±. 53 ±.5.7 ±. 5 ±.5.73 ± ±.5.74 ±. 57