Time Study of Harvesting Equipment Using Gps - Derived Positional Data

The objectives in this study were to develop and test a data analysis system for calculating machine productivity from GPS-derived positional information alone. A technique was used where positions were ‘filtered’ initially to locate specific events that were independent of what actually traveled the path, then these events were combined using user-specified rules into machine functions. The system was successful in producing gross-time-study da*& but less so in providing detailed elemental times. INTRODUCTION . Tie study is a basic tool used in studying the effects of management factors on productivity of logging systems. It has been used for many years in calculating costs for logging practices (Gardner 1963), and is fundamental in the analysis of forest operations. Collecting time study data, however, can be an expensive process (Olsen and Kellogg 1983). Time study of forest operations equipment ordinarily requires considerable travel to job sites. Forest operations tend to be dispersed across a large site, requiring several people to monitor activities scattered throughout the unit. There are many potential safety hazards working around woods equipmenf requiring well-trained, careful field crews. Although field coilection of data has been successful in the past, technology has evolved to the point that perhaps it ‘is time to consider replacing the risk and cost of field crews with autonomous data collection systems. Time study of forest harvesting systems common to the southern US is a relatively simple process. Especially for skidding, there are relatively few time elements to capture, and function is highly correlated with where the equipment is located on the site. Given the location of key areas, such as the deck or a delimbing gate, elemental times for skidding can be inferred almost entirely from where the skidder is relative to these positions. Automation of time study for these machines, therefore, may require only position data, which are simple and inexpensive to collect using Global Positioning Systems (GPS), and can be acquired remotely. Although skidding is the simplest case in measuring production from position, it is by no means the only one. Any method for autonomous production studies would have to be flexible enough to analyze a number of different types of machines performing several different functions. Many of these functions would likely be unrelated to any fried position. To fully capture sign&ant time elements of the production cycles, any autonomous time study system would need to be scalablein order to accept input data from a number of alternative sources. The purpose of this research was to investigate methods for continuous autonomous monitoring of harvest system productivity. Initial efforts have focused on the use of positional data alone to infer time study information for forest harvesting equipment. Successful development of such a system would provide a cost-effective means of studying the efficiency of harvesting systems for long periods of time, providing detailed insight into the effects of management, equipment, and site factors on harvest efficiency. Specific objectives of this phase of the study were: 1) Develop methods and software for analyzing the movements of forest harvesting equipment to calculate productivity. 2 ) Use the developed system to perform time study on a skidder. This paper describes the development and design objectives of a software analysis system for converting GPS-derived positional data into productivity infimnation. The data requirements for the system are outlined, and an illustration of its use in calculating skidder productivity is presented. EXPERIMENTAL METHODS Tie study of forest harvesting equipment normally involves breaking the work cycle of a given machine into discrete tasks, or elements. Each of these elements has a time of performance associated with it, and perhaps some other parameter, e.b. 0 distance traversed or volume of timber handled. In the field, elements are timed relative to specific starting and ending events. A skid cycle, for example, might be broken down into a (1) travel empty, (2) posirion and grapple, and (3) travel loaded sequence. Each of these elements has a characteristic event that marks its beginning and end: travel empty begins when the skidder leaves the deck and ends when the skidder stops prior to backing up to a load. ’ To perform time studies, an automated system would require some way of identifying the time and location of specific, individual events, and then be able to combine those events into sequences meaningful to the machine’s function. Our system for deriving time study information from positional data was implemented using two components: 1) a ‘feature’ extraction sub-system to identify basic events in a machine path, and 2) an ‘event processor’ 10 combine characteristic movements and sub-events into machine-specific functions. The approach used a small set of feantres measured corn a GPSderived path that were independent of what acmally created the motion to describe the function of a machine. Language recognition is a useful conceptual model for understanding the time study system development. The path itself was filtered, much as speech might be filtered, to produce a sequence of fundamental events, or ‘words’ in the language recoFition context. These events were then parsed using a specific grammar to form ‘sentences’, or sequences of events, describing the machine functionally. Users of the system decided what events, or ‘words’, were important to describe the function of a particular machine, then how those ‘words’ were arranged into meaningful combinations. A skidder represents perhaps the simplest example of how this approach can be used to derive time study information. The fimdamental events, or ‘words’, important in describing skidder function might include things like ‘leaves deck’, or ‘enters deck’, or ‘reverses direction’. Combinations of these words, e.g. ‘leaves deck; enters deck’ would represent more complicated stxuctures, in this case perhaps a ‘skid cycle’. The time study analysis system developed for this project was an implementation of this approach: a GPSderived path was filtered to extract important basic features of the path, then a user-defined grammar was used to further filter, or ‘reduce’, the fundamental events into machine functions. The two steps were implemented as two computer programs: (1) eventmap to filter a path and extract interesting events; and (2) eventparse to combine events into machine functions. To summarize the above, the process for reducing positional data to productivity information involved two stages of analysis. The first stage identified machine-independent features of the recorded motion, and the second collected those features into actions using a description of the machine’s ‘behavior’. These steps were implemented as two C++ programs, eventmap and eventparse, each requiring an input file with syntax outlined in the following. The analysis was completed when no more event parsing rules could be matched, after which the final list of events was output along with event times and total distance covered. The ‘fundamental events’ extracted fi-om the raw GPS data were from a set of measurable features that included the following: Enter or leave a polygon Occupy a location (Inside or outside a polygon) Cross a linear feature Start or stop moving Reverse direction Note that these features are independent of what actually created the path. This made it possible to derive the fundamental events for describing the actions of any machine using the same software tool. This software tool (eventmap) read as input a machine path, and a specification of which events to filter out and report on. The syntax for describing fundamental events to be identified was like the following: