3.1. Elemental Time Study
Over the four field campaigns (
Table 4), the trial crew was monitored for about 203 h in total. During that time, detailed time studies on 12 yarding corridors have been conducted, with a total of 687 yarding cycles valid for analysis. The net (m
3/PSH
0) and gross (m
3/SSH) productivity has been highest during campaign B, representing the campaign with the shortest corridors and also the lowest number of observations. In contrast, campaign C had the lowest net and gross output but owing to the terrain, also the longest corridors and lateral yarding distances. Campaign C also had the highest number of observations. In general, mean lateral yarding distances increased across the first three campaigns, reaching the highest mean value of 22.9 m in campaign C, as mentioned before. Yet, during the last campaign (D), the mean lateral yarding distance dropped down to 15.4 m.
In all four data sets, the
Hook Up process was the most time consuming cycle element, contributing to nearly one third of the total cycle time, also having the highest variability of the elements (
Appendix Table A1). With respect to machine utilization rates (
Appendix Table A2), MU
Total has been lowest during campaign A in 2013, but it also showed the highest MU
Operation. Campaign C had the lowest MU
Operation value and the highest DF
Total. However, it was also the longest observed period with an increased tendency of technical delays, as confirmed by the highest values for DF
MT and DF
RT (
Table A2).
Comparing the productive time (PSH
0) share to non-work times and supportive work times at shift level (SSH
Total), (
Figure 3) and at machine operation level (SSH
Operation), (
Figure 4) between the campaigns, the high share of PT (preparatory times) on shift level is noticeable, which is mainly due to the preparation process of the yarder and associated rigging. Rigging procedures attributed to corridor installation and take-down have been timed separately and evaluated by applying a suitable European corridor installation times model for small yarders defined by a mainline pull capacity <35 kN, as defined by Stampfer et al. [
33] (
Table A3).
We are aware of the limited possibility of comparing European conditions in spruce forests of the Austrian Alps, with their longer extraction distances when compared to Chinese pine plantations. But we consider Stampfer’s [
33] model appropriate for the purpose in order to compare the Chinese start-up crew with a professional one, particularly with respect to the rigging efforts of the intermediate supports and thereby suggesting a general performance level. It is obvious that during the first corridors in particular, the installation times took much longer than the modeled reference. The very first timed installation took as much as three times as the modelled reference. Later, the measured installation times do not show a clear pattern; installation and take down times being close to the estimated times (e.g., installation of corridor 5) but then at times it was much slower than expected, as observed during installation of corridor 7, while at others were much faster than expected, such as the installation of corridor 11 (
Table A3). Generally, the mean measured installation (4.14 h) and take down (1.94 h) times required 41.2% more time than predicted with Stampfer’s model [
33], yet a general improvement over time could not be identified.
The shift level times (SMH
Total) emphasize the high dependency of the yarding system on the loader, which if temporarily unavailable, forced the yarding operations to be placed on hold, even though the yarder itself was available for work. This was mainly due to the accumulation of logs at the deck which could not be removed manually and required the loader. Reasons for the unavailability have been technical break-downs, as for example shown by the high share of RT (repair time) of 27.7% (
Figure 3) on shift level times in campaign B, corresponding to a loss of 5.27 h production time. Other impacts were organizational deficits such as the occupation of the loader for truck loading during actual scheduled yarding hours as a result of uncoordinated trucking times. This represented for example, 12.9% or 2.25 h of IT (interference time) during field campaign A and therefore caused reductions in the overall potential productive times.
When focusing on machine operation only (SSH
Operation), the productive work time (PSH
0) has, as expected, the highest share of the overall times (
Figure 4), as also outlined by a high MU
Operation compared to MU
Total (
Table A2). ITs are dominating the SSH
Operation next to the productive work time among all four campaigns. Main cause of IT have been interruptions of the yarding process caused by hang ups but also frequent discussions between the yarder operator and choker setter on the radio, to clarify the work procedures and other work-related issues (
Table A4). In contrast to the SSH
Total, PTs only make up a small fraction amongst all campaigns but increased its share within the last two campaigns. They mainly consisted of building up pressure on the carriage as a necessity due to short yarding distances and tensioning of the skyline, which were both frequently occurring but only accounted for short lasting PTs.
Neither on an operational nor on shift level did MT (maintenance time) or RP (rest and personal times) play a major role.
Among the models computed for net-cycle times through stepwise multiple regressions, the campaign B data set model had the lowest fit, which explains only 36% of the observed variance (
Table 5). In contrast, the model of best fit explains 66% (
Table 5) of the variance and was generated for the campaign C data set. Campaign C had the highest number of observations but also the highest rate of disturbances through IT and PT on the machine operation times (18.9% and 6.1%, respectively) (
Figure 4) when compared to the other three campaigns. Although the number of explanatory variables differed between the campaigns, “Yarding Distance” and “Lateral Distance” are included in all models and not surprisingly, have significant effects on net-cycle times.
A standardization attempt was undertaken by using the individual cycle time equations (
Table 5) of the four campaigns and the overall mean values of all four campaigns (
Table 4) for the various independent variables. Under these assumptions, the observations during campaign C again revealed the shortest mean cycle times (5.10 min/cycle), followed by campaign D (5.44 min/cycle), campaign B (5.85 min/cycle) and campaign A (6.55 min/cycle). Plotting the standardized independent variables with the mean values over an increasing yarding distance up to 75 m, a yarding range covered by all studied corridors, gives a general confirmation of these indications (
Figure 5). Furthermore, the visualization shows that during the first campaign A, cycle times with associated short yarding distances were relatively short in duration but ascend faster with increasing yarding distances than during the other periods, particularly at ranges above 25 m.
The same approach has been used to illustrate the differences between the individual campaigns for the net
Outhaul,
Lateral Outhaul,
Lateral Inhaul and
Inhaul times, using the cycle element’s individual regression models (
Table A5) and the time demand dependent on yarding distance (up to 75 m) and lateral distance (up to 30 m) for the associated element (
Figure 5).
The individual models for
Hook Up and
Unhooking times (
Table A5) have no joint significant predictor that can conveniently display the effects of operational conditions such as yarding or lateral distance for
Outhaul and
Inhaul and
Lateral Outhaul and
Lateral Inhaul, respectively. Therefore, box plots (
Figure 6) have been used to visualize the variation in time demand between the campaigns of these cycle elements. The median values for
Hook Up and 50% of the respective observations show a lower time demand for campaign D, with a significant difference of the mean time (Df = 3,
p < 0.05) compared to the previous campaigns. Individual outliers (
Figure 6) can generally be associated to poor log presentation with slash or crossed over logs hampering the choker setting, but also due to individual logs located at ridges that were difficult to reach.
The relatively short times for
Unhooking compared to
Hook Up show higher time demand for campaign B and C and shorter demand for campaign A and D (
Figure 6). The difference of campaign A and D were both significant compared to B and C in terms of time demand (Df = 3,
p < 0.01). An abrupt increase of mean lateral yarding distances by 33% occurred from campaign A to B, while campaign C had almost similarly long mean lateral distances as campaign B (
Table 4). Consequently, a reduced lift effect was observed by the time keepers during the
Lateral Inhaul associated to the longer lateral distances. This caused the load to be dragged on the ground through the understory, twisting the chokers, taking up vegetation and slash which impeded the
Unhooking and led to delays such as slash hindering the
Unhooking process (see
Table A4). During the last field campaign in 2015 (campaign D), the lateral yarding distances had again been reduced and this time, the amount of co-extracted slash to the log deck resulted in a lower average time demand for
Unhooking when compared to campaigns B and C.
The use of standardized cycle times (as in
Figure 5) for calculation of comparative mean net-productivity would generate 6.65 m
3/PSH
0 for the most productive operation campaign C, followed by 6.23 m
3/PSH
0 for campaign D, 5.81 m
3/PSH
0 for campaign B and 5.18 m
3/PSH
0 for campaign A. When the net-productivity is displayed for increasing yarding distance of up to 75 m, which was covered within all corridors (
Figure 7), it shows relatively consistent parallel trends for campaigns C, D and B, similar to those of the pure cycle times. The productivity during campaign A was very high within the short distances, even superior to campaign C but the performance drops rapidly with increasing distances, being below that of all the other campaigns beyond the 37 m mark.
3.2. Log Book Records
Keeping a log book was a challenge for the cable yarding crew and initial recordings had to be discarded due to incomplete data and false classification of the individual times. After a technical review during field campaign B, useable recordings were achieved for the period from May 2014 to July 2015, covering 32 corridors and 2344 yarding cycles. This period equals 422 days, including 287 potential work days after deduction of weekends and regional holidays. Yet, actual work related to the yarding operation was scheduled and executed only on 197 days, amounting to a total yarded volume of 1392.4 m3, which represents a mean gross daily output of 7.1 m3. Heavy rainfalls impeding forest operations were the main cause of the reduction in work days; access to the forest also continued to be restricted in subsequent days due to poor forest road standards. Additional down days were caused by institutional deficits associated to selection of cut blocks and delays of issuing corresponding harvesting permits. Finally, deficits in timely communication of work progress effected delays in completion of planning and organizing work of subsequent operations.
There are 1122.1 SSH that were recorded in total (SSH
Total), which amount to a productive work time of 348.4 PSH
15, representing a MU
Total of 31.0%. The high share of PT (37.1%) (
Figure 8) is again mainly associated to rigging activities. Recorded rigging times and corridor specifications showed that the average corridor had a length of 113 m, was spaced at 44 m, had one intermediate support, required a total mean set-up time of 5.1 h and corresponding mean take-down time of 2.0 h. This is also covered by the ranges observed in the 12 corridors during the four field campaigns, although these long term average time demands are slightly longer. The share of IT (29.3%) (
Figure 8) sums up to a total of 200.4 h, of which 129.8 h were associated to truck loading by the loader. Logistical timing deficits were identified that resulted in the use of the knuckle boom loader, which is obligatory for efficient operation of the yarding system and for truck loading processes instead of clearing the log deck of the yarder; and by this, interrupting the yarder operation leading to avoidable reductions in PSH
15. Further cases in which IT reduced the potential MU
Total were: backlogs in tree felling, road construction and leveling of landings. As mentioned before, during the long term monitoring, IT contributed to 29% of the overall SSH
Total, nearly twice as much as the 15% average throughout the detailed time studies. In contrast, over the long-term the dominant time demand for PT, 37.1%, was generally shorter than that of the detailed time studies and made up approximately half of the SSH
Total, apart from campaign B, for which it amounted to 1/5 of SSH
Total (
Figure 3). RP times did not occur during the long-term records since short breaks and rests are already included within the PSH
15 due to the recordings of lesser detail.
From the log book recordings and calculated SSHTotal system utilization an overall performance estimate was derived. Considering 365 days per year, SSHTotal was generated by deducting weekends and local holidays resulting in 245 potential work days per year. Assuming eight hour daily shifts (working hours without accounting for break times), this sums up to 1960 SSHTotal/year, which relates to 608 PSH15 when considering a 31% MUTotal, as revealed by the log book. Based on the recorded log book performance of approximately 4 m3/PSH15 this results in a theoretical output of 2428 m3/year or 202 m3/month. By considering every potential work day during the studied period, the log book recordings revealed that in practice, a mean value of system employment of only 5.7 SSHTotal/day can be assigned. At the recorded productivity level of the long term records, an extrapolation of these figures on an annual level would result in 1396 SSHTotal, 433 PSH15 and a corresponding total annual productivity expectation of 1732 m3. Machine scheduling and cost estimations should however, actually be based on full working days. Therefore, it is more plausible to account for full work days with eight hour shifts and to apply a reduction factor that is based on the ratio of potential work days and actual observed work days during the long-term recordings, which in this case amounts to 0.59. Under these assumptions, 139 working days remain annually. If proper organization would allow for full eight SSHTotal shifts on each of these work days, the current annual capacity of the crew under local conditions would amount to 1112 SSHTotal hours with an output of 1379 m3 at a 31% MUTotal.
3.3. Efficiency Analysis by Stochastic Frontier Analysis (SFA)
Based on the identified independent variables for the cycle time equations, variables were chosen to define the inefficiency model (Equation (6)) for the technical inefficiency component
in the SFA. Since these variables are significantly affecting the cycle time demand, the input variables for the production function, Equation (4) but are not directly related to the skill level of the crew, they are a source of inefficiency affecting the crews’ EEF and thus can be accounted for by the SFA model (Equation (5)). This implies that the remaining deviation from the potential frontier isoquant can be related to the skill and performance level of the crew.
where:
u = is the nonnegative technical inefficiency component
z = is a parameter to be estimated
yd = is the yarding distance (m)
ld = is the lateral distance (m)
vegin = is the indicator for vegetation hindrance (yes/no)
lpin = is the indicator for the quality of log presentation (bad/good)
slin = is the indicator for terrain slope (>35%: yes/no)
The data originally collected for the 3-level factor variables and , as well as the numeric variable , were converted into binary (indicator) variables in order to simplify the inefficiency model. Where and indicates an effect above the original level 1 and considers the terrain to be steep above a gradient of 35%. Values “yes” and “bad” were used in the model as a dummy variable 1; otherwise 0 was used.
The SFA is summarized in
Table 6 with the corresponding maximum likelihood estimates of the stochastic frontier production function (Equation (5)) and the inefficiency model (Equation (6)).
The two variables of the production function, namely cycle time (T) and number of logs (L), have positive and highly significant coefficients, confirming their high contribution to output productivity in volume per turn. The elasticity of the coefficients assign at a cycle time increase by 1% an increase of the production output by 0.24% and an increase in number of logs by 1%, would increase the production output by 0.48%. The returns of scale, as the sum of the input coefficients is <1, affirming the production function having decreasing returns to scale. The scale effect (coefficient) of the number of logs on productivity (output) is higher compared to that of the cycle time. However, the piece size is also a highly influential factor when considering cycle output with the load volume in relation to the number of logs per cycle. It is obvious that if the log number exceeds the amount of the four available chokers, then the individual piece size generally decreases once more, since only smaller logs are choked together by one choker cable. This results in decreasing load volumes, particularly when eight logs are attached, although observed only twice (
Figure 9).
Among the variables determined for the inefficiency model, only slope showed a significant effect on efficiency (
Table 6), with the negative coefficient indicating that steeper slopes would decrease the inefficiency. The coefficient for
indicates that approximately 86% of the deviation from the ideal production is explained by technical inefficiency (
u) and that the noise (
v) is of minor influence. A likelihood ratio test as integral part of the package frontier [
32], was conducted in order to verify whether the inefficiency term
u significantly improves the fit of the frontier model by comparing the stochastic frontier model with the corresponding OLS model under the hypothesis H
0: γ = 0 (
Table 7). As H
0 is rejected, the presence of technical inefficiency is furthermore confirmed and a justification of the stochastic frontier model with the EEF established.
The mean efficiency of the crew for all assessed yarding cycles during the period monitored was determined to have an EEF value of 0.62, representing 62% of the potential achievable output at the given input resources. The efficiency levels within the yarding cycles ranged from 0.18 to 0.95, showing variability across the entire period with no pattern, indicating a steady learning process, and the crew not having achieved a working routine yet. Grouping the mean efficiency values of assessed yarding cycles for each individual corridor and according to time periods associated the four field campaigns (
Figure 10), it is noticeable that at every campaign, the crew usually began with a low mean efficiency but improved by the next corridor within the same campaign. The overall lowest mean efficiency of 0.59 was encountered during campaign A and it also included the lowest mean efficiency at corridor level of 0.49 for corridor 1, directly at the beginning of the entire campaign. A significant difference between the mean efficiency values of corridor 1 and 2 is indicated by the confidence intervals (95%) that do not overlap which was in contrast to corridor 2 and 3, which had different means but overlapping confidence intervals. In addition, the confidence intervals of the corridors of the following period, campaign B, do not show significant differences in their mean efficiency, neither within the campaign nor to the previous one. Campaign C, however, shows a significantly lower mean efficiency value for its first corridor (corridor 6), compared to the previous campaign but a significant improvement for every additional corridor of the period. Although small fluctuations are visible in campaign D, the mean efficiency values do not differ significantly among all the corridors, which leads to a high mean efficiency value of 0.65, the highest among all campaigns. The efficiency is relative consistent within campaign D and shows no significant difference between the first corridor of campaign D and the last one of campaign C. This could indicate a certain degree of performance stabilization and with an approach towards the learning plateau, also indicated by a significant difference (Kruskal Wallis test; Df = 3,
p = 0.02) between the overall mean efficiency value for campaign A (0.59) and campaign D (0.65) (
Figure 10).