Using st2smv to Analyze Schedules

Overview

st2smv can be used to check whether a given automation system will allow a proposed schedule of operations to proceed. This mode of operation assumes that the schedule defines a sequence of future tasks that are supposed to occur at specific times in the future, and that some of the transitions in the automation system have a known minimum amount of time associated with them.

The relevant command-line options are:

  • --read-timing-data, which reads a data file (specified using the --input option) and creates a corresponding file that defines the timing dynamics in SMV syntax
  • --check-schedule, which reads a schedule and the current state of the process (specified in that order using the --input option) along with the relevant dynamics (listed in a metadata file which is specified using the --metadata option)
  • --solver-path, which specifies the path to the solver executable to use when analyzing the schedule (SynthSMV is known to work, and NuSMV may also work)

A complete example showing how to use those options and the required format of the input files is included below.

Example

Consider a process that consists of three steps. Steps 1 and 3 always require a fixed amount of time to be completed, while the amount of time required to complete Step 2 depends on the outcome of Step 1.

(the following sequence of commands will run the entire example)

./create_model.sh
./create_timing.sh
./check-t0.sh
./check-t5-no_delay.sh
./check-t5-delay.sh

Automation Logic

The following PLC code (a function block, written in Structured Text) implements the step transitions and fault detection logic:

FUNCTION_BLOCK process

    VAR_INPUT
	move_to_next_step : BOOL;
	fault_signal : BOOL;
    END_VAR

    VAR
	step_0 : BOOL := TRUE;
	step_1 : BOOL := FALSE;
	step_2 : BOOL := FALSE;
	step_3 : BOOL := FALSE;
	fault_occurred_in_step_1 : BOOL := FALSE;
    END_VAR

BEGIN

    // Record any fault that occurs during Step 1.
    IF step_1 AND fault_signal THEN
	fault_occurred_in_step_1 := TRUE;
    END_IF;

    // Step transitions (0 -> 1 -> 2 -> 3 -> 0).
    IF step_0 AND move_to_next_step THEN
	step_0 := FALSE;
	step_1 := TRUE;
    ELSIF step_1 AND move_to_next_step THEN
	step_1 := FALSE;
	step_2 := TRUE;
    ELSIF step_2 AND move_to_next_step THEN
	step_2 := FALSE;
	step_3 := TRUE;
	fault_occurred_in_step_1 := FALSE; // reset the "fault status"
    ELSIF step_3 AND move_to_next_step THEN
	step_3 := FALSE;
	step_0 := TRUE;
    END_IF;

END_FUNCTION_BLOCK

An additional step (Step 0) is included to indicate that the process is waiting to start.

The following command converts the ST code to a model:

st2smv --convert --input process.st --output-directory dynamics

Timing

The following data file (written in JSON) lists the amount of time that each step takes:

{
    "process": {
	"step_0": 0,
	"step_1": 5,
	"step_2": {
	    "(! fault_occurred_in_step_1)": 2,
	    "fault_occurred_in_step_1": 4
	},
	"step_3": 1
    }
}
  • There is no minimum amount of time that the process is required to wait before starting, so a time of 0 is given for Step 0.
  • Steps 1 and 3 each have a fixed duration (of 5 minutes and 1 minute, respectively).
  • The duration of Step 2 depends on whether the fault_occurred_in_step_1 variable is true: if not, it takes 2 minutes, otherwise, it takes 4 minutes.

The following command converts the timing data to a set of constraints in SMV format, which will be used to augment the model of the automation logic:

st2smv --read-timing-data --input timing.json --output-directory dynamics

Completion of Tasks

Because the automation system doesn't explicitly track the completion of tasks, the following additional SMV code is used to track the number of tasks that have been completed (i.e., the number of times that the process has finished Step 3 and moved back to Step 0 to wait for the next task):

TRANS next(COUNT_process_task) = (
    step_3 & next(step_0)
    ? COUNT_process_task + 1
    : COUNT_process_task
);

Metadata

A metadata file (written in JSON) describes which of the dynamics pertain to which tasks:

{
    "*": [
	"dynamics/model.json",
	"dynamics/timing.smv"
    ],
    "process": {
	"task": ["counter-process_task.smv"]
    }
}

The special value * indicates dynamics which must always be included (in this case, the model of the automation logic itself and the timing information). The other values indicate task-specific dynamics — in this case, the counter that tracks the number of tasks that the process has completed.

A Feasible Schedule

In the following schedule (written in JSON), the process is scheduled to start a single task at time 0 and finish it at time 8 (the empty string after the task's start and end times is an optional label, which is simply left blank here):

{
    "horizon": [
	0,
	10
    ],
    "tasks": {
	"process": [
	    [
		"task",
		[
		    0,
		    8
		],
		""
	    ]
	]
    }
}

The following file (written in JSON) indicates the initial state of the process, which is in Step 0 at time 0:

{
    "step_0": true,
    "step_1": false,
    "step_2": false,
    "step_3": false,
    "fault_occurred_in_step_1": false,
    "_absolute_time": 0,
    "time": 0
}

The _absolute_time variable represents the amount of time that has passed since the beginning of the entire operation, and time represents the current time relative to the start of the current schedule; they are both 0 in this case, because the schedule begins at the start of the operation.

The following command (written as a shell script here, but easily adapted to other platforms) checks whether the schedule is feasible (i.e., can be completed on time) if the system is in the given state:

st2smv --check-schedule \
    --input schedule-t0.json state-t0.json \
    --metadata metadata.json \
    --output-directory results-t0 \
    --verbosity debug \
    --solver-path SynthSMV \
;

In this case, it is feasible, because the process can complete Steps 1, 2, and 3 in 5 minutes, 2 minutes, and 1 minute, respectively, for a total of 8 minutes, which meets the scheduled deadline. Note that this requires that the task start immediately and not spend any time waiting in Step 0, and it also requires that Step 2 finish in 2 minutes (not 4), meaning that no fault can occur in Step 1.

Another Feasible Schedule

After 5 minutes have passed and Step 1 is finished, the following partial schedule remains to be completed (5 minutes of the initial horizon and 3 minutes of the time allotted to the first task remain):

{
    "horizon": [
	0,
	5
    ],
    "tasks": {
	"process": [
	    [
		"task",
		[
		    0,
		    3
		],
		""
	    ]
	]
    }
}

If no fault occurred during Step 1, then the current state will be the following:

{
    "step_0": false,
    "step_1": false,
    "step_2": true,
    "step_3": false,
    "fault_occurred_in_step_1": false,
    "_absolute_time": 5,
    "time": 0
}

where _absolute_time is 5 (because 5 minutes have passed since the beginning of the operation) and the relative time is still 0, because the schedule has been updated to only include the remaining part of the original schedule.

The following command confirms that the schedule is still feasible:

st2smv --check-schedule \
    --input schedule-t5.json state-t5-no_delay.json \
    --metadata metadata.json \
    --output-directory results-t5_no_delay \
    --verbosity debug \
    --solver-path SynthSMV \
;

An Infeasible Schedule

If a fault did occur during Step 1, then the state would be the following:

{
    "step_0": false,
    "step_1": false,
    "step_2": true,
    "step_3": false,
    "fault_occurred_in_step_1": true,
    "_absolute_time": 5,
    "time": 0
}

where fault_occurred_in_step_1 is now true.

Checking the previous schedule against this state indicates that the task can no longer finish by the scheduled deadline:

st2smv --check-schedule \
    --input schedule-t5.json state-t5-delay.json \
    --metadata metadata.json \
    --output-directory results-t5_delay \
    --verbosity debug \
    --solver-path SynthSMV \
;

In this case, the task will require at least 5 more minutes to finish (4 minutes for Step 2, due to the fault during Step 1, and 1 minute for Step 3, as usual), which will run past the deadline of 3 minutes.

An output file (results-t5_delay/delay.json) is produced that contains the information about the delay, including the time at which it occurred and the minimum amount by which the task will be delayed (2 minutes, in this case).